- Email: [email protected]
Determining factors affecting cross-border transport on King Fahd Causeway and comparison with urban travel
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Case Studies on Transport Policy journal homepage: www.elsevier.com/locate/cstp
Determining factors affecting cross-border transport on King Fahd Causeway and comparison with urban travel ⁎
Uneb Gazdera, , Nedal T. Ratroutb a b
Department of Civil Engineering, University of Bahrain, Isa Town, Bahrain Department of Civil and Environmental Engineering, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
A R T I C LE I N FO
A B S T R A C T
Keywords: King Fahd Causeway Cross-border traffic Non-traffic parameters
Exogenous (non-traffic) factors, especially the economic conditions of the countries, may impact the cross-border travel demand. This study investigates the effects of non-traffic factors on border transport of the King Fahd Causeway. Moreover, a brief comparison is also provided with studies on urban traffic prediction. The historic stock market indices were used as an indicator of the economic and political conditions of Bahrain and Saudi Arabia which are joined by the Causeway. Stock market prices were found to have a statistically significant effect on the prediction models. Moreover, the other non-traffic parameters, including the weather (i.e. humidity) and the religious vacation period, also have a significant effect on the prediction models. Historic (time series) traffic data has been commonly used for urban travel. This is not the case for border travel where time series models may not be suitable because the changes in political and economic conditions may have an unprecedented change in travel demand. The use of economic parameters has not been found in the prediction studies for urban traffic. It was found that the model with only non-traffic parameters can predict daily traffic with approximately 85% accuracy on the validation datasets and can be used if time-series is not available or impractical to use.
1. Introduction Travel demand is often affected by the factors which are not directly related to traffic, such as weather, vacations, availability of disposable money (which occurs when monthly salary is released) and network reliability (Saneinejad et al., 2012; Mishra et al., 2018). In addition to that, travel demand shows a cyclic behavior with respect to time (De Jong et al., 2003). Cultures and other related parameters are also reported to increase the complexity of cross-border traffic prediction (Meixell and Gargeya, 2005). Parameters considered for traffic generation models may change depending upon the characteristics of origin (or destination) zone. This trend can be observed in the Institute of Transportation Engineers (ITE) manuals, which show different trip generation models and rates for various types of land uses and their sub-categories (Institute of Transportation Engineers, 2015). Considering this fact, the models for predicting border transport become a unique problem since a wide variety of land uses and travel purposes are combined in this stream of traffic. The economic situation of the country is considered an important parameter for determining border transport. These parameters are often changed due to changes in national policies (Abate et al., 2018). Gross Domestic Product (GDP) has been used as an indicator of
⁎
these changes for predicting border transport (Petersen, 2011). Alternately, stock market price is also assumed to be a suitable option in this regard as it is affected by the economic situation as well as the political stability of the country (Rogowski, 1989; Przeworski and Limongi, 1993). The effects of economic and political conditions are generally indistinguishable from one another. Stock market fluctuations are observed in a relatively short term as compared to GDP, which is beneficial for prediction of daily traffic. Furthermore, stock indices could be taken as surrogate indicators of many other relevant variables such as; change in economic policies, impact of political regime/condition and changes in trade relations with possible changes in commodity prices. Employing readily available stock indices instead of measuring all the above variables makes the modeling task easier. Changing traveler base and transportation facilities are also identified as other factors causing uncertainty in border transport forecasting (De Neufville, 2008). Traveler base has been referred to as the type and rate of population (Gramillano et al., 2016). These indicators are not observed on daily basis and hence cannot be used for predicting daily traffic. Moreover, economical and technological changes have also been identified as a primary indicator of cross-border transport (Medeiros, 2017). These changes have a direct impact on stock market indices and their effects can be quantified through these indices.
Corresponding author. E-mail addresses: [email protected] (U. Gazder), [email protected] (N.T. Ratrout).
https://doi.org/10.1016/j.cstp.2018.09.003 Received 28 March 2018; Received in revised form 6 September 2018; Accepted 8 September 2018 2213-624X/ © 2018 World Conference on Transport Research Society. Published by Elsevier Ltd. All rights reserved.
Please cite this article as: Gazder, U., Case Studies on Transport Policy, https://doi.org/10.1016/j.cstp.2018.09.003
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
Table 1 Input/output variables. Variable
Description
Input/Output
KSA_stock Bahrain_stock ADT_total traffic Total traffic Var-A Hajj Ramadan
Stock market prices in Saudi Arabia stock market, taken as fortnightly averages (1–15, 16–31) Stock market prices in Bahrain stock market, taken as fortnightly averages (1–15, 16–30/31) Weekly average traffic (vehicles per day) crossing the King Fahd Causeway in both directions Daily traffic (vehicles per day) crossing the King Fahd Causeway in both directions Dummy variable, taken as 1 for 24–1 of the month, showing the salary disbursement period in Saudi Arabia Dummy variable, taken as 1 for 1–10 of the 12th Islamic month (lunar calendar), which is a vacation period in Saudi Arabia and Bahrain Dummy variable, taken as 1 for 24 of the 9th Islamic month to 3rd of the 10th Islamic month, which is a vacation period in Saudi Arabia and Bahrain Average daily temperature for Dammam (Saudi Arabia) and Bahrain (°C) Average daily humidity for Dammam (Saudi Arabia) and Bahrain (%)
Input Input Output Output Input Input Input
Temperaturea Humiditya a
Input Input
Source: https://en.tutiempo.net/climate.html.
2.1. Study area
This study is aimed at forecasting traffic on King Fahd Causeway which carries cross border traffic between Kingdoms of Saudi Arabia and Bahrain. Keeping the above factors in mind, the stock market prices of Saudi Arabia and Bahrain were used to predict weekly Average Daily Traffic (ADT) for border transport between these two countries. The weekly ADT was predicted using stock market indices because the effect of these indices (political and economic situation) will usually take a few days and will not show on micro scale such as daily traffic. Following that, other non-traffic parameters, including weather, vacation and salary periods, and other time-related parameters, were used, with the predicted weekly ADT, in the model for predicting daily traffic. The focus of this research was to investigate the effects of readily available non-traffic parameters on border transport prediction in this region. These parameters may cause changes in travel demand, which will not be shown in the periodic trend of traffic. Moreover, the continuous collection of traffic data for long periods requires substantial resources and state-of-the-art technology for accurate traffic counts (Gramaglia et al., 2014). The data collection technologies are undergoing changes with time, but there is still a lack of integration between travel demand modelers and practice communities (Lee et al., 2016; Kisgyörgy and Vasvári, 2017). Furthermore, time-series data for predicting cross-border transport may not be viable since non-traffic parameters may cause unprecedented change in traffic pattern. Therefore, a prediction model based upon non-traffic parameters may be more convenient in this case. This study is inspired by a previous study, done in the same area, by El-Alfy et al. (2015). It is found to be a pioneering study on the use of stock market prices for border traffic forecasting. Two concerns were identified in the above mentioned study, i.e. use of stock market indices for predicting daily traffic and use of Artificial Neural Networks (ANNs). The effects of national economic and political situations, which is surrogated by stock prices, may not always be affecting travel demand on a daily basis. It seems more logical to expect that historic data of prices for a longer time span (week or fortnight) are beneficial in predicting traffic. Secondly, ANNs do not give insight about the relationship between the predicted value and its predictors. Regression models would be a better approach considering their explanatory powers established through statistical bases. These concerns have been addressed in this research. Moreover, time series traffic data was used in developing the prediction models by El-alfy et al. (2015) in their study, which is not feasible, as mentioned above. Hence this study makes use of readily available non-traffic parameters for predicting border traffic. This study also provides the comparison of urban and border traffic prediction parameters and approaches.
The study area comprises of King Fahd Causeway, which provides the only land link for travelers to/from Bahrain and Saudi Arabia. Both countries are part of the Gulf Coperation Council (GCC). Saudi Arabia is a major oil producing country, while Bahrain is an important tourism and business attraction in the region. Trips attracted to Bahrain is increasing with time due to its focus to promote tourism (Mansfeld and Winckler, 2008). Bahrain is also considered one of the lowest risk countries among the Arab and Asian countries by tourists from western countries (Morakabati et al., 2012). Furthermore, Bahrain airport acts as a regional hub for connecting international flights to many countries of the world. The causeway is more commonly used by travelers from GCC countries in the absence of ferry and railway service, and with predominantly car-oriented cultures (Coombe, 1985; Al-Atawi, 2005). Therefore, accurate estimates of the traffic on this causeway become important for its sustained operation and planning for any other alternate modes (rail and/or ferry) in the future. 2.2. Analysis The variables used in this study are listed in Table 1 along with their description. These variables have been selected based upon review of the previous literature as well as authors’ understanding of the travel demand on the causeway. Moreover, all these variables conform the objective of the study as they are readily available online through various sources. The traffic data was acquired from the King Fahd Causeway authority for the period from January 2001 to October 2011. During this period, fuel prices were stable in the region and consequently not considered as possible predictor of traffic between the two countries. Minimum value for traffic was approximately 10,000 veh/ day while maximum value reaches more than 35,000 veh/day with an average growth rate of approximately 25% per annum. Average daily traffic during the study period was 18,522 veh/day with a standard deviation of 5324 veh/day. Prediction models were developed in two steps: firstly, statistical models were developed for predicting weekly average daily traffic (ADT) using historical stock market prices of prior week(s). The effect of stock data was found to be reasonable as the coefficients for these variables in the regression model were statistically significant for the ADT of the immediate week and the ADT with a lag of one week. The fortnight average was used for stock market prices in this study because it has been found as a reasonable period to gauge the effects of economic changes on the stock market (Obaidullah, 1992; Draper and Paudyal, 2002). Then, the second step was to translate the average weekly predictions into daily predictions using non-traffic parameters. In this step, the regression models were developed using the non-traffic parameters only and then with the addition of historic traffic (previous day’s traffic). The predictions by each model were adjusted using the daily and monthly adjustment factors. These factors were calculated as per Eqs.
2. Methodology and dataset This section presents a brief description of the study area and the dataset used in this study. This will help the readers to understand the discussion and justification of the results presented in the paper. 2
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout n
Table 2 Regression models with non-traffic parameters.
MAPE =
∑ i=1
Factor
Value
p-value
Predicting Weekly ADT Intercept Saudi Arabia_stock No. of data points (n) Adjusted R-square RMSE MAPE Pearson’s correlation between observed and predicted traffic
15966.41 0.45 237 0.13 4082.06 19.63% 0.37
2.6E−74 2.7E−09
Predicting Daily Traffic Intercept Ramadan Hajj Humidity Variable A ADT_total traffic_predicted Adjusted R-square RMSE MAPE Pearson’s Correlation between observed and predicted traffic
623.57 −2023.62 2707.59 −16.83 520.41 1.00 0.12 4621.43 22.58% 0.36
0.64 5.42E−04 1.76E−06 0.03 0.04 2.60E−48
(1)
Adjusted traffic = (predicted daily traffic)/(daily adjustment factor (3)
Overall ADT refers to the mean of all values in the data sample, while ADT for day of the week refers to the mean of all values for specific day of the week. For example, daily adjustment factor for Monday is calculated as the ratio of overall ADT and the ADT for Mondays only. Similarly, monthly adjustment was calculated as the overall ADT divided by the ADT of the specific month of the year. For example, ADT for January is the mean value of all values for January in the dataset for the complete study period. Such adjustment factors are used to convert ADT into daily traffic or vice versa in traffic studies (Roess et al., 2011). The adjustment factors may account for any seasonal variations in traffic. The approach to predict an average value of traffic and then further refine it to get the exact prediction has also been used for predicting urban traffic (Lippi et al., 2013) and passenger flow on urban rail (Feng et al., 2016a,b). The accuracies of these models have been reported in terms of Pearson’s correlation coefficient (R), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) and coefficient of determination (R-square) parameter. Correlation coefficient indicates the change in two variables with respect to each other. It ranges from −1 to 1, where values in the proximity of 0 indicate a weak relationship, 1 indicates a positively strong relationship (increase in a variable corresponds to increase in other) and −1 indicates negatively strong relationship (decrease in a variable corresponds to increase in other). Rsquare parameter indicates the proportion of variance captured by the model, it ranges from 0 to 1. These parameters are calculated by Eqs. (4)–(6) (Chou and Pham, 2013).
3.1. Statistical analysis Based upon the above-mentioned observations, it was decided to use the historic fortnightly average Saudi Arabia stock market price for predicting the future weekly average traffic on the causeway for the immediately succeeding week (without any lag). The regression model developed between these parameters is shown in Table 2. Although the stock market has a low correlation with traffic, the probability (p-value) for its coefficient in the regression model was found to be approximately ‘0’, which shows that the coefficient for stock market price is statistically different than ‘0’, and thus it has a significant impact on the model. The regression model was able to predict with an average error of approximately 20%, which amounts to 4000 vehicles per day. So it can be said that the model gives 80% accurate results which could be considered reasonable in the absence of comprehensive historical traffic data or if the use of historic data is not feasible as it could be the case for border transport. The predicted weekly ADT, In addition to other non-traffic parameters (as mentioned in Table 1), except Bahrain stock market prices, were used to develop a regression model for predicting the daily traffic on the causeway. The results of the regression equation are shown in Table 2. Only those variables were included in the final model which had coefficients with p-value of 5% or less. The intercept for the model was found to be statistically insignificant but it was kept in the model since omitting it would force the model to the origin which is impractical. It would mean that if the predicting variables considered in the study are zero then there would be zero traffic which is not usually true. Furthermore, regression models, which are forced to origin, are not considered directly comparable to other regression models, especially for R-square comparisons (Eisenhauer, 2003; Neter et al., 1996). It can be observed from Table 2 that the predicted border traffic is
n
n
n
n ∑i = 1 yi . yi '−( ∑i = 1 yi )( ∑i = 1 yi ') n
n
2
n
n
'
n ( ∑i = 1 yi 2 −( ∑i = 1 yi ) n ( ∑i = 1 yi'2 −(∑i = 1 yi )2
RMSE =
1 n
(4)
n
∑ (yi −yi' )2 i=1
(6)
First of all, Pearson’s correlation between Bahrain and Saudi Arabia stock markets was calculated, which showed that these stock markets are highly correlated with a correlation coefficient of 0.82. The possible reason for this high correlation of Bahrain and Saudi Arabia stock market prices could be the fact that the economic conditions in Saudi Arabia also boost/degrade the economy of Bahrain. The high correlation coefficient also implies that only one of these stock prices could be used in the regression equation, hence, the second step was to find the correlation between these stock market prices and the traffic. It was found that Saudi Arabia has a higher correlation with traffic as compared to Bahrain (0.37 vs. 0.27). This could be because the majority of the traffic on this corridor is generated from Saudi Arabia for tourism and business purposes due to Bahrain’s recreational attraction points and flexible banking and financial systems (Khan and Bhatti, 2008; Nuruzzaman, 2013). This traffic would be more susceptible to changes in the economic situation of the Saudi Arabia stock market. Therefore, it was decided to use the Saudi Arabia stock market prices for predicting the weekly average traffic. Correlation between fortnightly average Saudi Arabia stock market price and the weekly average traffic was calculated for the immediate week and that with a lag of one week. The correlation coefficient was relatively higher (0.37) for the weekly average traffic of the immediate week as compared to that with one week lag (0.25). This observation is consistent with the results achieved with ANN models in an earlier study on this dataset where lag in the input and output variables reduced the accuracy of the models (El-Alfy et al., 2015).
(2)
R=
× 100
3. Analysis and discussion
Monthly adjustment factor = (overall ADT)/(ADT for month of the year)
× monthly adjustment factor)
yi
where y′ is the predicted travel demand, y is the actual travel demand, and n is the number of data samples.
(1)–(3).
Daily adjustment factor = (overall ADT)/(ADT for day of the week)
|yi −yi' |
(5) 3
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
1.2 Monthly adjustment factor
affected positively by Hajj vacations and salary period (indicated by variable A). The reason for this could be that people from Saudi Arabia often travel to Bahrain for recreational and social reasons, which is more common during vacations and when they receive the salary. Ramadan vacation has a negative effect on the model because Muslims (which is majority of population in Saudi Arabia and Bahrain) fast during most of these vacations and would not like to travel for recreational purposes. The coefficient of humidity was found to be affecting the model significantly with a negative sign, while the temperature was eliminated because its coefficient was statistically close to ‘0’. The possible reason for that could be that the temperature in both countries is normally high for most days of the year with less variation, having a standard deviation of approximately 7 °C. On the other hand, humidity varies more in different seasons of the year, having a standard deviation of around 14%. The ADT predicted through stock market prices was still found to be significantly affecting this model. Therefore, the use of political and economic conditions (surrogated by stock market prices in predicting ADT) of Saudi Arabia is justified for prediction border transport on King Fahd Causeway. The signs (positive and negative) in this model satisfy general understanding about the travel behavior for this border travel. Adjustment factors were calculated for each day of the week and each month of the year using Eqs. (1) and (2). Figs. 1 and 2 present a comparison of daily and monthly adjustment factors calculated from the available data. The data presented in Fig. 1 shows that Thursday has the lowest adjustment factor, indicating the relatively higher volume of traffic on this day. The most obvious reason for this could be the fact that Thursday was the start of the weekend in both countries during the study period (i.e. 2001–2011) and, as seen in the regression model, the traffic demand is higher during the holiday/vacation period. Fig. 2 shows that January, December, June, and July have higher traffic demand. These months often include the end of semester breaks in educational institutes in both countries when families would travel more. The predicted traffic was adjusted using these factors as shown in Eq. (3). The performance measures calculated for the observed and the predicted traffic, after adjustment, are given in Table 3. The use of adjustment factors resulted in improving the accuracy of the predictions by 3% for MAPE and increased the Pearson’s correlation between the predicted and observed traffic by 0.18. The above-mentioned process was repeated for predicting daily traffic with the addition of previous day’s traffic as the independent variable in the regression model. To ascertain that all the input variables are independent of each other, correlation was found between the predicted weekly ADT and the previous day traffic and it was found to be 0.30. This can be considered as practically insignificant and therefore the condition of independence is satisfied. Redeveloping the regression model, with previous day traffic, resulted in omission of some factors which were not found to be significantly affecting the traffic in
0.8 0.6 0.4 0.2 0
Fig. 2. Comparison of monthly adjustment factors.
Table 3 Performance measures for adjusted traffic. Factor
Value
RMSE MAPE Pearson’s Correlation between observed and predicted traffic
4178.66 19.26% 0.54
Table 4 Regression model and performance measures for predicting daily traffic with previous day’s traffic. Factor
Value
p-value
Intercept Hajj ADT_total traffic_predicted (t-1) daily traffic Adjusted R-square RMSE MAPE Pearson’s Correlation between observed and predicted traffic
217.06 698.35 0.11 0.87 0.78 2307.46 9.64% 0.88
0.74 0.01 0.00 0.00
Performance after applying adjustment factors RMSE MAPE Pearson’s Correlation between observed and predicted traffic
2248.20 9.59% 0.90
the presence of its historic value. The regression equation along with the performance measures for prediction with and without adjustment factors is shown in Table 4. It can be observed in this case that the inclusion of the previous day’s traffic has resulted in the exclusion of most of the exogenous variables. However, the effect of political and economic conditions, included in the form of ADT predicted by the stock market prices, still has a significant effect on the daily traffic predictions. Similar to Table 2, the coefficient for the intercept was found to be statistically insignificant but it was kept in the model due to the reasons mentioned previously in the discussion of Table 2. The use of the adjustment factors in this case did not have a considerable change in the accuracy of the predictions. King Fahd Causeway authority website has made available the yearly data for 2012 and 2013, including the total annual traffic and ADT for each month. The daily data for any period after 2011 could not be found. So, the above approach of using non-traffic parameters and adjustment factors was validated by predicting the daily traffic for each day of 2012 and 2013. Then, the total yearly and monthly averages were calculated using these predicted values for comparison with the official data. The performance measures for this validation are shown in Table 5. It should be noted that the annual total traffic has a large magnitude, so the RMSE value has also been found to be relatively higher. However, the relative accuracy measures like MAPE and Rsquare values present the validation in a better manner. Both these
1.2 Daily adjustment factor
1
1 0.8 0.6 0.4 0.2 0
Fig. 1. Comparison of daily adjustment factors. 4
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
need for this study to use non-traffic parameters for border traffic prediction. The approach of this study is different from that adopted for urban travel demand modeling, wherein the characteristics of the area, travelers, and their household are mainly considered as the factors affecting travel demand (Al-Atawi, 2005; Motuba and Tolliver, 2017). However, it could not be done in this case, since the travel corridor includes travelers from a country of varied characteristics rather than a small community of similar characteristics. In addition, number of studies were found in which urban traffic has been forecasted using ANNs, regression models, and time series methods. Most of these studies have focused on short-term traffic forecasting using time series data (Smith and Demetsky, 1997; Dia, 2001; Vlahogianni et al., 2004; Vlahogianni et al., 2014; Wu et al., 2014; Xu et al., 2016). Tsirigotis et al. (2012) have also found that using exogenous factors such as weather, only marginally improves the traffic prediction accuracy. There have been some commonalities found between the prediction of urban traffic and the present study which are as follows. In some of the studies, use of cluster analysis and pattern recognition has also been supported to increase the accuracy of prediction models by incorporating seasonal variations and time of the day (Danech-Pajouh and Aron, 1991; Van Der Voort et al., 1996; Cai et al., 2016; West and Börjesson, 2018). The same has also been adopted in this study with the use of daily and monthly adjustment factors. However, it was found in the present study that the use of these factors was not beneficial for predictions when historical traffic values were used in making them, which was not the case in the above-mentioned studies. The use of incident occurrences (accidents, rainfall, etc.) has also been advocated in some of the studies (Chen and Grant-Muller, 2001; Tsirigotis et al., 2012). For cross border travel, these incidents could be related to political and economic instability in the country. This reinforces the use of stock market prices to incorporate the effects of economic and political incidents on the traffic. This is because measuring the extent and duration of the effects of these events is rather difficult for the crossborder travel. Similar to the present study, the effects of weather (including temperature and visibility among other parameters) have also been found in the study conducted by Chen et al. (2014). There are some parameters which have been used for urban traffic forecasting and seem applicable to border transport as well. There is a need to incorporate the network effects on traffic predictions as being identified in the case of urban travel, wherein the effects of traffic/ congestion on one link (upstream/downstream) have been observed to have effects on the other links/segments as well (Wu et al., 2014; Vlahogianni, 2015). Other such effects include the introduction of alternate transport services, accessibility to different modes, and changes in highway network (Feng et al., 2016a,b; Huang et al., 2017). Employment opportunities and access to urban transport systems could also be possible indicators of border transport (Gramillano et al., 2016). These parameters could be helpful if a route provides access to multiple countries which is not the case in this study.
Table 5 Performance measures for validation dataset. Validation dataset
Factor
Value
Monthly average traffic
RMSE MAPE
4226.72 15.13%
Annual total traffic
RMSE MAPE
1204136.21 13.62%
parameters seem to be reasonable, taking into consideration the fact that they are calculated for the monthly ADTs and annual total traffic. They are better than the accuracies obtained for the dataset used in the modeling, which could be because the economic and political conditions are having more effect on traffic demand on the causeway recently. Hence, using the ADT, predicted by the stock market prices, proves to be more beneficial for the relatively recent data. In comparison with a previous study for this study area by El-Alfy et al. (2015), the use of only non-traffic parameters was found to have resulted in reducing the accuracy by 7%. This difference can be considered acceptable since the models in this study are more practical (by using readily available non-traffic parameters) and employing regression models which are intutivelly appealing and lend additional explanatory power to the predictions. The R-square values were observed to be very low for the regression model with non-traffic parameters. Jin et al. (2001) also observed low values of R-square for regression models as observed in their study. They pointed out that R-square values are sensitive to the irregularity of the problem as shown in Eq. (7).
R2 = 1−
(Mean square error for predicted and observed values ) Variance of observed data (7)
Glosten et al. (1993) advocated to not use R-square parameter for measuring accuracy for models with the dummy variables. On the basis of these observations, it can be said that the use of RMSE and MAPE are better measures for determining the accuracy of these models in the case of non-traffic parameters which includes a number of dummy variables. Furthermore, the value of R-square is also dependent upon the mean square error to some extent, which is included as one of the performance measures. Hence, focusing on RMSE instead of R-square for evaluating the accuracy of the model is more appealing in this study. 3.2. Differences and similarities with other regions Morakabati et al. (2012) also conducted a study to investigate the factors affecting the travelers’ choice of traveling to Middle East countries from UK. Their findings, based upon subjective responses of the travelers, confirm the finding of this study that political and economic conditions do affect the travelers’ choice to travel to a specific destination. Moreover, they also found that weather is also an important parameter for these travelers, which has also been the case in this study. Moreover, the salary period dummy variable has affected the travel demand on the causeway in this study, which can also be related to the economic conditions of the travelers. Hence, it can be said that the travel from western countries is mostly affected by the same parameters as that of the travel from neighboring countries in the study region. Economic-technical parameters have also been found as the primary indicator for border transport among Scandanavian countries as well (Medeiros, 2017). From the review of cross border transport prediction literature, it can be observed that the border transport is considered to be more unique than urban traffic and shows less tendency to follow specific patterns (Lin et al., 2014). Hence, the use of time series models have not been applied to border traffic prediction as observed in the studies by Petersen (2011) and Morakabati et al. (2012). These studies justify the
4. Conclusions This study was focused on investigating the effects of non-traffic parameters on cross-border traffic on King Fahd Causeway. Moreover, it also presents a comparative review of the findings of this study with other borders and urban traffic prediction studies. Regression models were used in this study and their independent variables included weather parameters (temperature and humidity), dummy variables for salary and vacation periods, and stock market prices. The historic stock market prices were used as an indicator of the economic and political volatility in the country which could change due to any prior event. It was observed that the border transport has a relatively higher correlation with the economic and political conditions in the Saudi Arabia. Humidity was found to be the most important weather 5
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
parameter which could be attributed to its higher degree of variation in different seasons in the study area. The traffic demand increases during the vacation periods (Hajj, summer and winter breaks) and weekly holidays as shown by the model and the weekly and monthly adjustment factors. Use of historic traffic data in the prediction model increases the accuracy of the model and reduces the number of independent variables. However, the effect of political and economic situations (inserted via predicted weekly ADT predicted using stock market prices) was still found to have a significant effect on the model. Lastly, the use of readily available non-traffic parameters in the prediction model provided an accuracy of approximately 85–80% (i.e. 15–20% MAPE) in predicting cross border daily traffic values. This may be useful in this case since the time-series models may not be applicable. The comparison with other previous studies also confirmed that political and economic situations (for border transport), similar to any significant incidences on the network (for urban travel), can affect the traffic. It was also found that weather and seasonal variations in travel demand had a significant effect on traffic predictions for urban as well as border traffic. However, the use of economic parameters has not been found useful for urban traffic predictions while it is used regularly for border traffic prediction. In the context of future research, it is recommended to investigate the effects, of other modes of transportation (airplane) and traffic demand on other borders of Bahrain and Saudi Arabia, on the causeway traffic.
rationally forecasting traffic congestion probability. Int. J. Prod. Res. 54 (12), 3465–3474. Feng, X., Zhang, H., Gan, T., Sun, Q., Ma, F., Sun, X., 2016b. Random coefficient modeling research on short-term forecast of passenger flow into an urban rail transit station. Transport 31 (1), 94–99. Glosten, L.R., Jagannathan, R., Runkle, D.E., 1993. On the relation between the expected value and the volatility of the nominal excess return on stocks. J. Finance 48 (5), 1779–1801. Gramaglia, M., Calderon, M., Bernardos, C.J., 2014. ABEONA monitored traffic: VANETassisted cooperative traffic congestion forecasting. IEEE Veh. Technol. Mag. 9 (2), 50–57. Gramillano A., Levarlet F., Nilsson H., Camagni R., Capello R., Caragliu A., et al., 2016. Collecting solid evidence to assess the needs to be addressed by Interreg cross-border cooperation programmes. Directorate General for Regional and Urban Policy. Huang, Z., Zhang, M., Liu, X., 2017. Estimating light-rail transit peak-hour boarding based on accessibility at station and route levels in Wuhan, China. Transp. Plan. Technol. 40 (5), 624–639. Institute of Transportation Engineers (ITE)., 2015. Trip Generation Manual, 9th edn., Washington, DC 20006 USA. Jin, R., Chen, W., Simpson, T.W., 2001. Comparative studies of metamodelling techniques under multiple modelling criteria. Struct. Multidiscip. Optim. 23 (1), 1–13. Khan, M.M., Bhatti, M.I., 2008. Islamic banking and finance: on its way to globalization. Manag. Finance 34 (10), 708–725. Kisgyörgy, L., Vasvári, G., 2017. Concept of an open-access travel demand modeling platform. Case Stud. Transp. Policy 5 (3), 453–459. Lee, R.J., Sener, I.N., Mullins III, J.A., 2016. An evaluation of emerging data collection technologies for travel demand modeling: From research to practice. Transp. Lett. 8 (4), 181–193. Lin, L., Wang, Q., Huang, S., Sadek, A.W., 2014. On-line prediction of border crossing traffic using an enhanced spinning network method. Transp. Res. Part C: Emerg. Technol. 43, 158–173. Lippi, M., Bertini, M., Frasconi, P., 2013. Short-term traffic flow forecasting: An experimental comparison of time-series analysis and supervised learning. IEEE Trans. Intell. Transp. Syst. 14 (2), 871–882. Mansfeld, Y., Winckler, O., 2008. The role of the tourism industry in transforming a rentier to a long-term viable economy: The case of Bahrain. Curr. Issues Tour. 11 (3), 237–267. Medeiros, E., 2017. Cross-border cooperation in inner Scandinavia: A territorial impact assessment. Environ. Impact Assess. Rev. 62, 147–157. Meixell, M.J., Gargeya, V.B., 2005. Global supply chain design: A literature review and critique. Transp. Res. Part E: Logist. Transp. Rev. 41 (6), 531–550. Mishra, S., Tang, L., Ghader, S., Mahapatra, S., Zhang, L., 2018. Estimation and valuation of travel time reliability for transportation planning applications. Case Stud. Transp. Policy 6 (2018), 51–62. Morakabati, Y., Fletcher, J., Prideaux, B., 2012. Tourism development in a difficult environment: a study of consumer attitudes, travel risk perceptions and the termination of demand. Tour. Econ. 18 (5), 953–969. Motuba, D., Tolliver, D., 2017. Truck trip generation in small-and medium-sized urban areas. Transp. Plan. Technol. 40 (3), 327–339. Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1996. Applied Linear Statistical Models, vol. 4 McGraw-Hill/Irwin, Chicago. Nuruzzaman, M., 2013. Politics, economics and Saudi military intervention in Bahrain. J. Contemp. Asia 43 (2), 363–378. Obaidullah, M., 1992. How do stock prices react to bonus issues? Vikalpa 17 (1), 17–22. Petersen, T., 2011. Modelling cross-border transport in Öresund: Scenario assumptions, model results and a before-and-after database. Rev. Reg. Res. (Jahrbuch für Regionalwissenschaft) 173–210. Przeworski, A., Limongi, F., 1993. Political regimes and economic growth. J. Econ. Perspect. 7 (3), 51–69. Roess, R.P., Prassas, E.S., McShane, W.R., 2011. Traffic Engineering, 4th ed. Prentice Hall, Upper Saddle River, N.J. Rogowski, R., 1989. Commerce and Coalitions: How Trade Affects Domestic Political Alignments, Princeton, N.J. Saneinejad, S., Matthew, J.R., Kennedy, C., 2012. Modelling the impact of weather conditions on active transportation travel behavior. Transp. Res. Part D: Transp. Environ. 17 (2), 129–137. Smith, B.L., Demetsky, M.J., 1997. Traffic flow forecasting: comparison of modeling approaches. J. Transp. Eng. 123 (4), 261–266. Tsirigotis, L., Vlahogianni, E.I., Karlaftis, M.G., 2012. Does information on weather affect the performance of short-term traffic forecasting models? Int. J. Intell. Transp. Syst. Res. 10 (1), 1–10. Van Der Voort, M., Dougherty, M., Watson, S., 1996. Combining Kohonen maps with ARIMA time series models to forecast traffic flow. Transp. Res. Part C: Emerg. Technol. 4 (5), 307–318. Vlahogianni, E.I., 2015. Optimization of traffic forecasting: Intelligent surrogate modeling. Transp. Res. Part C: Emerg. Technol. 55, 14–23. Vlahogianni, E.I., Golias, J.C., Karlaftis, M.G., 2004. Short-term traffic forecasting: Overview of objectives and methods. Transp. Rev. 24 (5), 533–557. Vlahogianni, E.I., Karlaftis, M.G., Golias, J.C., 2014. Short-term traffic forecasting: Where we are and where we’re going. Transp. Res. Part C: Emerg. Technol. 43, 3–19. West, J., Börjesson, M., 2018. The Gothenburg congestion charges: cost–benefit analysis and distribution effects. Transportation. https://doi.org/10.1007/s11116-0179853-4. Wu, S., Yang, Z., Zhu, X., Yu, B., 2014. Improved k-nn for short-term traffic forecasting using temporal and spatial information. J. Transp. Eng. 140 (7), 04014026. Xu, C., Li, Z., Wang, W., 2016. Short-term traffic flow prediction using a methodology based on autoregressive integrated moving average and genetic programming. Transport 31 (3), 343–358.
Acknowledgements The authors would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum and Minerals (KFUPM), Saudi Arabia for funding this work under Grant No. IN121065. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.cstp.2018.09.003. References Abate, M., Vierth, I., Karlsson, R., de Jong, G., Baak, J.A., 2018. Disaggregate stochastic freight transport model for Sweden. Transportation. https://doi.org/10.1007/ s11116-018-9856-9. Al-Atawi, A.M., 2005. Household Travel and Tour-based Behaviour: A Case Study from Saudi Arabia, Doctoral Dissertation, Cardiff University. Cai, P., Wang, Y., Lu, G., Chen, P., Ding, C., Sun, J., 2016. A spatiotemporal correlative knearest neighbor model for short-term traffic multistep forecasting. Transp. Res. Part C: Emerg. Technol. 62, 21–34. Chen, H., Grant-Muller, S., 2001. Use of sequential learning for short-term traffic flow forecasting. Transp. Res. Part C: Emerg. Technol. 9 (5), 319–336. Chen, X.Y., Pao, H.K., Lee, Y.J., 2014. Efficient traffic speed forecasting based on massive heterogenous historical data. In: 2014 IEEE Int. Conf. Big Data, pp. 10–17. Chou, J.S., Pham, A.D., 2013. Enhanced artificial intelligence for ensemble approach to predicting high performance concrete compressive strength. Constr. Build. Mater. 49, 554–563. Coombe, R.D., 1985. Urban transport policy development: two case studies in the Middle East. Transp. Rev. 5 (2), 165–188. Danech-Pajouh, M., Aron, M., 1991. ATHENA: a method for short-term inter-urban motorway traffic forecasting. Recherche Transports Sécurité 6, 6–11. De Jong, G., Daly, A., Pieters, M., Vellay, C., Bradley, M., Hofma, F., 2003. A model for time of day and mode choice using error components logit. Transp. Res. Part E: Logist Transp. Rev. 39 (3), 245–268. De Neufville, R., 2008. Low-cost airports for low-cost airlines: flexible design to manage the risks. Transp. Plan. Technol. 31 (1), 35–68. Dia, H., 2001. An object-oriented neural network approach to short-term traffic forecasting. Eur. J. Operat. Res. 131 (2), 253–261. Draper, P., Paudyal, K., 2002. Explaining Monday returns. J. Financ. Res. 25 (4), 507–520. Eisenhauer, J.G., 2003. Regression through the origin. Teach. Stat. 25 (3), 76–80. El-Alfy, E.S.M., Ratrout, N.T., Gazder, U., 2015. Impact of stock market indices and other regional exogenous factors on predictive modeling of border traffic with neural network models. Arab. J. Sci. Eng. 40 (2), 303–312. Feng, X., Saito, M., Liu, Y., 2016a. Improve urban passenger transport management by
6
Contents lists available at ScienceDirect
Case Studies on Transport Policy journal homepage: www.elsevier.com/locate/cstp
Determining factors affecting cross-border transport on King Fahd Causeway and comparison with urban travel ⁎
Uneb Gazdera, , Nedal T. Ratroutb a b
Department of Civil Engineering, University of Bahrain, Isa Town, Bahrain Department of Civil and Environmental Engineering, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
A R T I C LE I N FO
A B S T R A C T
Keywords: King Fahd Causeway Cross-border traffic Non-traffic parameters
Exogenous (non-traffic) factors, especially the economic conditions of the countries, may impact the cross-border travel demand. This study investigates the effects of non-traffic factors on border transport of the King Fahd Causeway. Moreover, a brief comparison is also provided with studies on urban traffic prediction. The historic stock market indices were used as an indicator of the economic and political conditions of Bahrain and Saudi Arabia which are joined by the Causeway. Stock market prices were found to have a statistically significant effect on the prediction models. Moreover, the other non-traffic parameters, including the weather (i.e. humidity) and the religious vacation period, also have a significant effect on the prediction models. Historic (time series) traffic data has been commonly used for urban travel. This is not the case for border travel where time series models may not be suitable because the changes in political and economic conditions may have an unprecedented change in travel demand. The use of economic parameters has not been found in the prediction studies for urban traffic. It was found that the model with only non-traffic parameters can predict daily traffic with approximately 85% accuracy on the validation datasets and can be used if time-series is not available or impractical to use.
1. Introduction Travel demand is often affected by the factors which are not directly related to traffic, such as weather, vacations, availability of disposable money (which occurs when monthly salary is released) and network reliability (Saneinejad et al., 2012; Mishra et al., 2018). In addition to that, travel demand shows a cyclic behavior with respect to time (De Jong et al., 2003). Cultures and other related parameters are also reported to increase the complexity of cross-border traffic prediction (Meixell and Gargeya, 2005). Parameters considered for traffic generation models may change depending upon the characteristics of origin (or destination) zone. This trend can be observed in the Institute of Transportation Engineers (ITE) manuals, which show different trip generation models and rates for various types of land uses and their sub-categories (Institute of Transportation Engineers, 2015). Considering this fact, the models for predicting border transport become a unique problem since a wide variety of land uses and travel purposes are combined in this stream of traffic. The economic situation of the country is considered an important parameter for determining border transport. These parameters are often changed due to changes in national policies (Abate et al., 2018). Gross Domestic Product (GDP) has been used as an indicator of
⁎
these changes for predicting border transport (Petersen, 2011). Alternately, stock market price is also assumed to be a suitable option in this regard as it is affected by the economic situation as well as the political stability of the country (Rogowski, 1989; Przeworski and Limongi, 1993). The effects of economic and political conditions are generally indistinguishable from one another. Stock market fluctuations are observed in a relatively short term as compared to GDP, which is beneficial for prediction of daily traffic. Furthermore, stock indices could be taken as surrogate indicators of many other relevant variables such as; change in economic policies, impact of political regime/condition and changes in trade relations with possible changes in commodity prices. Employing readily available stock indices instead of measuring all the above variables makes the modeling task easier. Changing traveler base and transportation facilities are also identified as other factors causing uncertainty in border transport forecasting (De Neufville, 2008). Traveler base has been referred to as the type and rate of population (Gramillano et al., 2016). These indicators are not observed on daily basis and hence cannot be used for predicting daily traffic. Moreover, economical and technological changes have also been identified as a primary indicator of cross-border transport (Medeiros, 2017). These changes have a direct impact on stock market indices and their effects can be quantified through these indices.
Corresponding author. E-mail addresses: [email protected] (U. Gazder), [email protected] (N.T. Ratrout).
https://doi.org/10.1016/j.cstp.2018.09.003 Received 28 March 2018; Received in revised form 6 September 2018; Accepted 8 September 2018 2213-624X/ © 2018 World Conference on Transport Research Society. Published by Elsevier Ltd. All rights reserved.
Please cite this article as: Gazder, U., Case Studies on Transport Policy, https://doi.org/10.1016/j.cstp.2018.09.003
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
Table 1 Input/output variables. Variable
Description
Input/Output
KSA_stock Bahrain_stock ADT_total traffic Total traffic Var-A Hajj Ramadan
Stock market prices in Saudi Arabia stock market, taken as fortnightly averages (1–15, 16–31) Stock market prices in Bahrain stock market, taken as fortnightly averages (1–15, 16–30/31) Weekly average traffic (vehicles per day) crossing the King Fahd Causeway in both directions Daily traffic (vehicles per day) crossing the King Fahd Causeway in both directions Dummy variable, taken as 1 for 24–1 of the month, showing the salary disbursement period in Saudi Arabia Dummy variable, taken as 1 for 1–10 of the 12th Islamic month (lunar calendar), which is a vacation period in Saudi Arabia and Bahrain Dummy variable, taken as 1 for 24 of the 9th Islamic month to 3rd of the 10th Islamic month, which is a vacation period in Saudi Arabia and Bahrain Average daily temperature for Dammam (Saudi Arabia) and Bahrain (°C) Average daily humidity for Dammam (Saudi Arabia) and Bahrain (%)
Input Input Output Output Input Input Input
Temperaturea Humiditya a
Input Input
Source: https://en.tutiempo.net/climate.html.
2.1. Study area
This study is aimed at forecasting traffic on King Fahd Causeway which carries cross border traffic between Kingdoms of Saudi Arabia and Bahrain. Keeping the above factors in mind, the stock market prices of Saudi Arabia and Bahrain were used to predict weekly Average Daily Traffic (ADT) for border transport between these two countries. The weekly ADT was predicted using stock market indices because the effect of these indices (political and economic situation) will usually take a few days and will not show on micro scale such as daily traffic. Following that, other non-traffic parameters, including weather, vacation and salary periods, and other time-related parameters, were used, with the predicted weekly ADT, in the model for predicting daily traffic. The focus of this research was to investigate the effects of readily available non-traffic parameters on border transport prediction in this region. These parameters may cause changes in travel demand, which will not be shown in the periodic trend of traffic. Moreover, the continuous collection of traffic data for long periods requires substantial resources and state-of-the-art technology for accurate traffic counts (Gramaglia et al., 2014). The data collection technologies are undergoing changes with time, but there is still a lack of integration between travel demand modelers and practice communities (Lee et al., 2016; Kisgyörgy and Vasvári, 2017). Furthermore, time-series data for predicting cross-border transport may not be viable since non-traffic parameters may cause unprecedented change in traffic pattern. Therefore, a prediction model based upon non-traffic parameters may be more convenient in this case. This study is inspired by a previous study, done in the same area, by El-Alfy et al. (2015). It is found to be a pioneering study on the use of stock market prices for border traffic forecasting. Two concerns were identified in the above mentioned study, i.e. use of stock market indices for predicting daily traffic and use of Artificial Neural Networks (ANNs). The effects of national economic and political situations, which is surrogated by stock prices, may not always be affecting travel demand on a daily basis. It seems more logical to expect that historic data of prices for a longer time span (week or fortnight) are beneficial in predicting traffic. Secondly, ANNs do not give insight about the relationship between the predicted value and its predictors. Regression models would be a better approach considering their explanatory powers established through statistical bases. These concerns have been addressed in this research. Moreover, time series traffic data was used in developing the prediction models by El-alfy et al. (2015) in their study, which is not feasible, as mentioned above. Hence this study makes use of readily available non-traffic parameters for predicting border traffic. This study also provides the comparison of urban and border traffic prediction parameters and approaches.
The study area comprises of King Fahd Causeway, which provides the only land link for travelers to/from Bahrain and Saudi Arabia. Both countries are part of the Gulf Coperation Council (GCC). Saudi Arabia is a major oil producing country, while Bahrain is an important tourism and business attraction in the region. Trips attracted to Bahrain is increasing with time due to its focus to promote tourism (Mansfeld and Winckler, 2008). Bahrain is also considered one of the lowest risk countries among the Arab and Asian countries by tourists from western countries (Morakabati et al., 2012). Furthermore, Bahrain airport acts as a regional hub for connecting international flights to many countries of the world. The causeway is more commonly used by travelers from GCC countries in the absence of ferry and railway service, and with predominantly car-oriented cultures (Coombe, 1985; Al-Atawi, 2005). Therefore, accurate estimates of the traffic on this causeway become important for its sustained operation and planning for any other alternate modes (rail and/or ferry) in the future. 2.2. Analysis The variables used in this study are listed in Table 1 along with their description. These variables have been selected based upon review of the previous literature as well as authors’ understanding of the travel demand on the causeway. Moreover, all these variables conform the objective of the study as they are readily available online through various sources. The traffic data was acquired from the King Fahd Causeway authority for the period from January 2001 to October 2011. During this period, fuel prices were stable in the region and consequently not considered as possible predictor of traffic between the two countries. Minimum value for traffic was approximately 10,000 veh/ day while maximum value reaches more than 35,000 veh/day with an average growth rate of approximately 25% per annum. Average daily traffic during the study period was 18,522 veh/day with a standard deviation of 5324 veh/day. Prediction models were developed in two steps: firstly, statistical models were developed for predicting weekly average daily traffic (ADT) using historical stock market prices of prior week(s). The effect of stock data was found to be reasonable as the coefficients for these variables in the regression model were statistically significant for the ADT of the immediate week and the ADT with a lag of one week. The fortnight average was used for stock market prices in this study because it has been found as a reasonable period to gauge the effects of economic changes on the stock market (Obaidullah, 1992; Draper and Paudyal, 2002). Then, the second step was to translate the average weekly predictions into daily predictions using non-traffic parameters. In this step, the regression models were developed using the non-traffic parameters only and then with the addition of historic traffic (previous day’s traffic). The predictions by each model were adjusted using the daily and monthly adjustment factors. These factors were calculated as per Eqs.
2. Methodology and dataset This section presents a brief description of the study area and the dataset used in this study. This will help the readers to understand the discussion and justification of the results presented in the paper. 2
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout n
Table 2 Regression models with non-traffic parameters.
MAPE =
∑ i=1
Factor
Value
p-value
Predicting Weekly ADT Intercept Saudi Arabia_stock No. of data points (n) Adjusted R-square RMSE MAPE Pearson’s correlation between observed and predicted traffic
15966.41 0.45 237 0.13 4082.06 19.63% 0.37
2.6E−74 2.7E−09
Predicting Daily Traffic Intercept Ramadan Hajj Humidity Variable A ADT_total traffic_predicted Adjusted R-square RMSE MAPE Pearson’s Correlation between observed and predicted traffic
623.57 −2023.62 2707.59 −16.83 520.41 1.00 0.12 4621.43 22.58% 0.36
0.64 5.42E−04 1.76E−06 0.03 0.04 2.60E−48
(1)
Adjusted traffic = (predicted daily traffic)/(daily adjustment factor (3)
Overall ADT refers to the mean of all values in the data sample, while ADT for day of the week refers to the mean of all values for specific day of the week. For example, daily adjustment factor for Monday is calculated as the ratio of overall ADT and the ADT for Mondays only. Similarly, monthly adjustment was calculated as the overall ADT divided by the ADT of the specific month of the year. For example, ADT for January is the mean value of all values for January in the dataset for the complete study period. Such adjustment factors are used to convert ADT into daily traffic or vice versa in traffic studies (Roess et al., 2011). The adjustment factors may account for any seasonal variations in traffic. The approach to predict an average value of traffic and then further refine it to get the exact prediction has also been used for predicting urban traffic (Lippi et al., 2013) and passenger flow on urban rail (Feng et al., 2016a,b). The accuracies of these models have been reported in terms of Pearson’s correlation coefficient (R), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) and coefficient of determination (R-square) parameter. Correlation coefficient indicates the change in two variables with respect to each other. It ranges from −1 to 1, where values in the proximity of 0 indicate a weak relationship, 1 indicates a positively strong relationship (increase in a variable corresponds to increase in other) and −1 indicates negatively strong relationship (decrease in a variable corresponds to increase in other). Rsquare parameter indicates the proportion of variance captured by the model, it ranges from 0 to 1. These parameters are calculated by Eqs. (4)–(6) (Chou and Pham, 2013).
3.1. Statistical analysis Based upon the above-mentioned observations, it was decided to use the historic fortnightly average Saudi Arabia stock market price for predicting the future weekly average traffic on the causeway for the immediately succeeding week (without any lag). The regression model developed between these parameters is shown in Table 2. Although the stock market has a low correlation with traffic, the probability (p-value) for its coefficient in the regression model was found to be approximately ‘0’, which shows that the coefficient for stock market price is statistically different than ‘0’, and thus it has a significant impact on the model. The regression model was able to predict with an average error of approximately 20%, which amounts to 4000 vehicles per day. So it can be said that the model gives 80% accurate results which could be considered reasonable in the absence of comprehensive historical traffic data or if the use of historic data is not feasible as it could be the case for border transport. The predicted weekly ADT, In addition to other non-traffic parameters (as mentioned in Table 1), except Bahrain stock market prices, were used to develop a regression model for predicting the daily traffic on the causeway. The results of the regression equation are shown in Table 2. Only those variables were included in the final model which had coefficients with p-value of 5% or less. The intercept for the model was found to be statistically insignificant but it was kept in the model since omitting it would force the model to the origin which is impractical. It would mean that if the predicting variables considered in the study are zero then there would be zero traffic which is not usually true. Furthermore, regression models, which are forced to origin, are not considered directly comparable to other regression models, especially for R-square comparisons (Eisenhauer, 2003; Neter et al., 1996). It can be observed from Table 2 that the predicted border traffic is
n
n
n
n ∑i = 1 yi . yi '−( ∑i = 1 yi )( ∑i = 1 yi ') n
n
2
n
n
'
n ( ∑i = 1 yi 2 −( ∑i = 1 yi ) n ( ∑i = 1 yi'2 −(∑i = 1 yi )2
RMSE =
1 n
(4)
n
∑ (yi −yi' )2 i=1
(6)
First of all, Pearson’s correlation between Bahrain and Saudi Arabia stock markets was calculated, which showed that these stock markets are highly correlated with a correlation coefficient of 0.82. The possible reason for this high correlation of Bahrain and Saudi Arabia stock market prices could be the fact that the economic conditions in Saudi Arabia also boost/degrade the economy of Bahrain. The high correlation coefficient also implies that only one of these stock prices could be used in the regression equation, hence, the second step was to find the correlation between these stock market prices and the traffic. It was found that Saudi Arabia has a higher correlation with traffic as compared to Bahrain (0.37 vs. 0.27). This could be because the majority of the traffic on this corridor is generated from Saudi Arabia for tourism and business purposes due to Bahrain’s recreational attraction points and flexible banking and financial systems (Khan and Bhatti, 2008; Nuruzzaman, 2013). This traffic would be more susceptible to changes in the economic situation of the Saudi Arabia stock market. Therefore, it was decided to use the Saudi Arabia stock market prices for predicting the weekly average traffic. Correlation between fortnightly average Saudi Arabia stock market price and the weekly average traffic was calculated for the immediate week and that with a lag of one week. The correlation coefficient was relatively higher (0.37) for the weekly average traffic of the immediate week as compared to that with one week lag (0.25). This observation is consistent with the results achieved with ANN models in an earlier study on this dataset where lag in the input and output variables reduced the accuracy of the models (El-Alfy et al., 2015).
(2)
R=
× 100
3. Analysis and discussion
Monthly adjustment factor = (overall ADT)/(ADT for month of the year)
× monthly adjustment factor)
yi
where y′ is the predicted travel demand, y is the actual travel demand, and n is the number of data samples.
(1)–(3).
Daily adjustment factor = (overall ADT)/(ADT for day of the week)
|yi −yi' |
(5) 3
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
1.2 Monthly adjustment factor
affected positively by Hajj vacations and salary period (indicated by variable A). The reason for this could be that people from Saudi Arabia often travel to Bahrain for recreational and social reasons, which is more common during vacations and when they receive the salary. Ramadan vacation has a negative effect on the model because Muslims (which is majority of population in Saudi Arabia and Bahrain) fast during most of these vacations and would not like to travel for recreational purposes. The coefficient of humidity was found to be affecting the model significantly with a negative sign, while the temperature was eliminated because its coefficient was statistically close to ‘0’. The possible reason for that could be that the temperature in both countries is normally high for most days of the year with less variation, having a standard deviation of approximately 7 °C. On the other hand, humidity varies more in different seasons of the year, having a standard deviation of around 14%. The ADT predicted through stock market prices was still found to be significantly affecting this model. Therefore, the use of political and economic conditions (surrogated by stock market prices in predicting ADT) of Saudi Arabia is justified for prediction border transport on King Fahd Causeway. The signs (positive and negative) in this model satisfy general understanding about the travel behavior for this border travel. Adjustment factors were calculated for each day of the week and each month of the year using Eqs. (1) and (2). Figs. 1 and 2 present a comparison of daily and monthly adjustment factors calculated from the available data. The data presented in Fig. 1 shows that Thursday has the lowest adjustment factor, indicating the relatively higher volume of traffic on this day. The most obvious reason for this could be the fact that Thursday was the start of the weekend in both countries during the study period (i.e. 2001–2011) and, as seen in the regression model, the traffic demand is higher during the holiday/vacation period. Fig. 2 shows that January, December, June, and July have higher traffic demand. These months often include the end of semester breaks in educational institutes in both countries when families would travel more. The predicted traffic was adjusted using these factors as shown in Eq. (3). The performance measures calculated for the observed and the predicted traffic, after adjustment, are given in Table 3. The use of adjustment factors resulted in improving the accuracy of the predictions by 3% for MAPE and increased the Pearson’s correlation between the predicted and observed traffic by 0.18. The above-mentioned process was repeated for predicting daily traffic with the addition of previous day’s traffic as the independent variable in the regression model. To ascertain that all the input variables are independent of each other, correlation was found between the predicted weekly ADT and the previous day traffic and it was found to be 0.30. This can be considered as practically insignificant and therefore the condition of independence is satisfied. Redeveloping the regression model, with previous day traffic, resulted in omission of some factors which were not found to be significantly affecting the traffic in
0.8 0.6 0.4 0.2 0
Fig. 2. Comparison of monthly adjustment factors.
Table 3 Performance measures for adjusted traffic. Factor
Value
RMSE MAPE Pearson’s Correlation between observed and predicted traffic
4178.66 19.26% 0.54
Table 4 Regression model and performance measures for predicting daily traffic with previous day’s traffic. Factor
Value
p-value
Intercept Hajj ADT_total traffic_predicted (t-1) daily traffic Adjusted R-square RMSE MAPE Pearson’s Correlation between observed and predicted traffic
217.06 698.35 0.11 0.87 0.78 2307.46 9.64% 0.88
0.74 0.01 0.00 0.00
Performance after applying adjustment factors RMSE MAPE Pearson’s Correlation between observed and predicted traffic
2248.20 9.59% 0.90
the presence of its historic value. The regression equation along with the performance measures for prediction with and without adjustment factors is shown in Table 4. It can be observed in this case that the inclusion of the previous day’s traffic has resulted in the exclusion of most of the exogenous variables. However, the effect of political and economic conditions, included in the form of ADT predicted by the stock market prices, still has a significant effect on the daily traffic predictions. Similar to Table 2, the coefficient for the intercept was found to be statistically insignificant but it was kept in the model due to the reasons mentioned previously in the discussion of Table 2. The use of the adjustment factors in this case did not have a considerable change in the accuracy of the predictions. King Fahd Causeway authority website has made available the yearly data for 2012 and 2013, including the total annual traffic and ADT for each month. The daily data for any period after 2011 could not be found. So, the above approach of using non-traffic parameters and adjustment factors was validated by predicting the daily traffic for each day of 2012 and 2013. Then, the total yearly and monthly averages were calculated using these predicted values for comparison with the official data. The performance measures for this validation are shown in Table 5. It should be noted that the annual total traffic has a large magnitude, so the RMSE value has also been found to be relatively higher. However, the relative accuracy measures like MAPE and Rsquare values present the validation in a better manner. Both these
1.2 Daily adjustment factor
1
1 0.8 0.6 0.4 0.2 0
Fig. 1. Comparison of daily adjustment factors. 4
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
need for this study to use non-traffic parameters for border traffic prediction. The approach of this study is different from that adopted for urban travel demand modeling, wherein the characteristics of the area, travelers, and their household are mainly considered as the factors affecting travel demand (Al-Atawi, 2005; Motuba and Tolliver, 2017). However, it could not be done in this case, since the travel corridor includes travelers from a country of varied characteristics rather than a small community of similar characteristics. In addition, number of studies were found in which urban traffic has been forecasted using ANNs, regression models, and time series methods. Most of these studies have focused on short-term traffic forecasting using time series data (Smith and Demetsky, 1997; Dia, 2001; Vlahogianni et al., 2004; Vlahogianni et al., 2014; Wu et al., 2014; Xu et al., 2016). Tsirigotis et al. (2012) have also found that using exogenous factors such as weather, only marginally improves the traffic prediction accuracy. There have been some commonalities found between the prediction of urban traffic and the present study which are as follows. In some of the studies, use of cluster analysis and pattern recognition has also been supported to increase the accuracy of prediction models by incorporating seasonal variations and time of the day (Danech-Pajouh and Aron, 1991; Van Der Voort et al., 1996; Cai et al., 2016; West and Börjesson, 2018). The same has also been adopted in this study with the use of daily and monthly adjustment factors. However, it was found in the present study that the use of these factors was not beneficial for predictions when historical traffic values were used in making them, which was not the case in the above-mentioned studies. The use of incident occurrences (accidents, rainfall, etc.) has also been advocated in some of the studies (Chen and Grant-Muller, 2001; Tsirigotis et al., 2012). For cross border travel, these incidents could be related to political and economic instability in the country. This reinforces the use of stock market prices to incorporate the effects of economic and political incidents on the traffic. This is because measuring the extent and duration of the effects of these events is rather difficult for the crossborder travel. Similar to the present study, the effects of weather (including temperature and visibility among other parameters) have also been found in the study conducted by Chen et al. (2014). There are some parameters which have been used for urban traffic forecasting and seem applicable to border transport as well. There is a need to incorporate the network effects on traffic predictions as being identified in the case of urban travel, wherein the effects of traffic/ congestion on one link (upstream/downstream) have been observed to have effects on the other links/segments as well (Wu et al., 2014; Vlahogianni, 2015). Other such effects include the introduction of alternate transport services, accessibility to different modes, and changes in highway network (Feng et al., 2016a,b; Huang et al., 2017). Employment opportunities and access to urban transport systems could also be possible indicators of border transport (Gramillano et al., 2016). These parameters could be helpful if a route provides access to multiple countries which is not the case in this study.
Table 5 Performance measures for validation dataset. Validation dataset
Factor
Value
Monthly average traffic
RMSE MAPE
4226.72 15.13%
Annual total traffic
RMSE MAPE
1204136.21 13.62%
parameters seem to be reasonable, taking into consideration the fact that they are calculated for the monthly ADTs and annual total traffic. They are better than the accuracies obtained for the dataset used in the modeling, which could be because the economic and political conditions are having more effect on traffic demand on the causeway recently. Hence, using the ADT, predicted by the stock market prices, proves to be more beneficial for the relatively recent data. In comparison with a previous study for this study area by El-Alfy et al. (2015), the use of only non-traffic parameters was found to have resulted in reducing the accuracy by 7%. This difference can be considered acceptable since the models in this study are more practical (by using readily available non-traffic parameters) and employing regression models which are intutivelly appealing and lend additional explanatory power to the predictions. The R-square values were observed to be very low for the regression model with non-traffic parameters. Jin et al. (2001) also observed low values of R-square for regression models as observed in their study. They pointed out that R-square values are sensitive to the irregularity of the problem as shown in Eq. (7).
R2 = 1−
(Mean square error for predicted and observed values ) Variance of observed data (7)
Glosten et al. (1993) advocated to not use R-square parameter for measuring accuracy for models with the dummy variables. On the basis of these observations, it can be said that the use of RMSE and MAPE are better measures for determining the accuracy of these models in the case of non-traffic parameters which includes a number of dummy variables. Furthermore, the value of R-square is also dependent upon the mean square error to some extent, which is included as one of the performance measures. Hence, focusing on RMSE instead of R-square for evaluating the accuracy of the model is more appealing in this study. 3.2. Differences and similarities with other regions Morakabati et al. (2012) also conducted a study to investigate the factors affecting the travelers’ choice of traveling to Middle East countries from UK. Their findings, based upon subjective responses of the travelers, confirm the finding of this study that political and economic conditions do affect the travelers’ choice to travel to a specific destination. Moreover, they also found that weather is also an important parameter for these travelers, which has also been the case in this study. Moreover, the salary period dummy variable has affected the travel demand on the causeway in this study, which can also be related to the economic conditions of the travelers. Hence, it can be said that the travel from western countries is mostly affected by the same parameters as that of the travel from neighboring countries in the study region. Economic-technical parameters have also been found as the primary indicator for border transport among Scandanavian countries as well (Medeiros, 2017). From the review of cross border transport prediction literature, it can be observed that the border transport is considered to be more unique than urban traffic and shows less tendency to follow specific patterns (Lin et al., 2014). Hence, the use of time series models have not been applied to border traffic prediction as observed in the studies by Petersen (2011) and Morakabati et al. (2012). These studies justify the
4. Conclusions This study was focused on investigating the effects of non-traffic parameters on cross-border traffic on King Fahd Causeway. Moreover, it also presents a comparative review of the findings of this study with other borders and urban traffic prediction studies. Regression models were used in this study and their independent variables included weather parameters (temperature and humidity), dummy variables for salary and vacation periods, and stock market prices. The historic stock market prices were used as an indicator of the economic and political volatility in the country which could change due to any prior event. It was observed that the border transport has a relatively higher correlation with the economic and political conditions in the Saudi Arabia. Humidity was found to be the most important weather 5
Case Studies on Transport Policy xxx (xxxx) xxx–xxx
U. Gazder, N.T. Ratrout
parameter which could be attributed to its higher degree of variation in different seasons in the study area. The traffic demand increases during the vacation periods (Hajj, summer and winter breaks) and weekly holidays as shown by the model and the weekly and monthly adjustment factors. Use of historic traffic data in the prediction model increases the accuracy of the model and reduces the number of independent variables. However, the effect of political and economic situations (inserted via predicted weekly ADT predicted using stock market prices) was still found to have a significant effect on the model. Lastly, the use of readily available non-traffic parameters in the prediction model provided an accuracy of approximately 85–80% (i.e. 15–20% MAPE) in predicting cross border daily traffic values. This may be useful in this case since the time-series models may not be applicable. The comparison with other previous studies also confirmed that political and economic situations (for border transport), similar to any significant incidences on the network (for urban travel), can affect the traffic. It was also found that weather and seasonal variations in travel demand had a significant effect on traffic predictions for urban as well as border traffic. However, the use of economic parameters has not been found useful for urban traffic predictions while it is used regularly for border traffic prediction. In the context of future research, it is recommended to investigate the effects, of other modes of transportation (airplane) and traffic demand on other borders of Bahrain and Saudi Arabia, on the causeway traffic.
rationally forecasting traffic congestion probability. Int. J. Prod. Res. 54 (12), 3465–3474. Feng, X., Zhang, H., Gan, T., Sun, Q., Ma, F., Sun, X., 2016b. Random coefficient modeling research on short-term forecast of passenger flow into an urban rail transit station. Transport 31 (1), 94–99. Glosten, L.R., Jagannathan, R., Runkle, D.E., 1993. On the relation between the expected value and the volatility of the nominal excess return on stocks. J. Finance 48 (5), 1779–1801. Gramaglia, M., Calderon, M., Bernardos, C.J., 2014. ABEONA monitored traffic: VANETassisted cooperative traffic congestion forecasting. IEEE Veh. Technol. Mag. 9 (2), 50–57. Gramillano A., Levarlet F., Nilsson H., Camagni R., Capello R., Caragliu A., et al., 2016. Collecting solid evidence to assess the needs to be addressed by Interreg cross-border cooperation programmes. Directorate General for Regional and Urban Policy. Huang, Z., Zhang, M., Liu, X., 2017. Estimating light-rail transit peak-hour boarding based on accessibility at station and route levels in Wuhan, China. Transp. Plan. Technol. 40 (5), 624–639. Institute of Transportation Engineers (ITE)., 2015. Trip Generation Manual, 9th edn., Washington, DC 20006 USA. Jin, R., Chen, W., Simpson, T.W., 2001. Comparative studies of metamodelling techniques under multiple modelling criteria. Struct. Multidiscip. Optim. 23 (1), 1–13. Khan, M.M., Bhatti, M.I., 2008. Islamic banking and finance: on its way to globalization. Manag. Finance 34 (10), 708–725. Kisgyörgy, L., Vasvári, G., 2017. Concept of an open-access travel demand modeling platform. Case Stud. Transp. Policy 5 (3), 453–459. Lee, R.J., Sener, I.N., Mullins III, J.A., 2016. An evaluation of emerging data collection technologies for travel demand modeling: From research to practice. Transp. Lett. 8 (4), 181–193. Lin, L., Wang, Q., Huang, S., Sadek, A.W., 2014. On-line prediction of border crossing traffic using an enhanced spinning network method. Transp. Res. Part C: Emerg. Technol. 43, 158–173. Lippi, M., Bertini, M., Frasconi, P., 2013. Short-term traffic flow forecasting: An experimental comparison of time-series analysis and supervised learning. IEEE Trans. Intell. Transp. Syst. 14 (2), 871–882. Mansfeld, Y., Winckler, O., 2008. The role of the tourism industry in transforming a rentier to a long-term viable economy: The case of Bahrain. Curr. Issues Tour. 11 (3), 237–267. Medeiros, E., 2017. Cross-border cooperation in inner Scandinavia: A territorial impact assessment. Environ. Impact Assess. Rev. 62, 147–157. Meixell, M.J., Gargeya, V.B., 2005. Global supply chain design: A literature review and critique. Transp. Res. Part E: Logist. Transp. Rev. 41 (6), 531–550. Mishra, S., Tang, L., Ghader, S., Mahapatra, S., Zhang, L., 2018. Estimation and valuation of travel time reliability for transportation planning applications. Case Stud. Transp. Policy 6 (2018), 51–62. Morakabati, Y., Fletcher, J., Prideaux, B., 2012. Tourism development in a difficult environment: a study of consumer attitudes, travel risk perceptions and the termination of demand. Tour. Econ. 18 (5), 953–969. Motuba, D., Tolliver, D., 2017. Truck trip generation in small-and medium-sized urban areas. Transp. Plan. Technol. 40 (3), 327–339. Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1996. Applied Linear Statistical Models, vol. 4 McGraw-Hill/Irwin, Chicago. Nuruzzaman, M., 2013. Politics, economics and Saudi military intervention in Bahrain. J. Contemp. Asia 43 (2), 363–378. Obaidullah, M., 1992. How do stock prices react to bonus issues? Vikalpa 17 (1), 17–22. Petersen, T., 2011. Modelling cross-border transport in Öresund: Scenario assumptions, model results and a before-and-after database. Rev. Reg. Res. (Jahrbuch für Regionalwissenschaft) 173–210. Przeworski, A., Limongi, F., 1993. Political regimes and economic growth. J. Econ. Perspect. 7 (3), 51–69. Roess, R.P., Prassas, E.S., McShane, W.R., 2011. Traffic Engineering, 4th ed. Prentice Hall, Upper Saddle River, N.J. Rogowski, R., 1989. Commerce and Coalitions: How Trade Affects Domestic Political Alignments, Princeton, N.J. Saneinejad, S., Matthew, J.R., Kennedy, C., 2012. Modelling the impact of weather conditions on active transportation travel behavior. Transp. Res. Part D: Transp. Environ. 17 (2), 129–137. Smith, B.L., Demetsky, M.J., 1997. Traffic flow forecasting: comparison of modeling approaches. J. Transp. Eng. 123 (4), 261–266. Tsirigotis, L., Vlahogianni, E.I., Karlaftis, M.G., 2012. Does information on weather affect the performance of short-term traffic forecasting models? Int. J. Intell. Transp. Syst. Res. 10 (1), 1–10. Van Der Voort, M., Dougherty, M., Watson, S., 1996. Combining Kohonen maps with ARIMA time series models to forecast traffic flow. Transp. Res. Part C: Emerg. Technol. 4 (5), 307–318. Vlahogianni, E.I., 2015. Optimization of traffic forecasting: Intelligent surrogate modeling. Transp. Res. Part C: Emerg. Technol. 55, 14–23. Vlahogianni, E.I., Golias, J.C., Karlaftis, M.G., 2004. Short-term traffic forecasting: Overview of objectives and methods. Transp. Rev. 24 (5), 533–557. Vlahogianni, E.I., Karlaftis, M.G., Golias, J.C., 2014. Short-term traffic forecasting: Where we are and where we’re going. Transp. Res. Part C: Emerg. Technol. 43, 3–19. West, J., Börjesson, M., 2018. The Gothenburg congestion charges: cost–benefit analysis and distribution effects. Transportation. https://doi.org/10.1007/s11116-0179853-4. Wu, S., Yang, Z., Zhu, X., Yu, B., 2014. Improved k-nn for short-term traffic forecasting using temporal and spatial information. J. Transp. Eng. 140 (7), 04014026. Xu, C., Li, Z., Wang, W., 2016. Short-term traffic flow prediction using a methodology based on autoregressive integrated moving average and genetic programming. Transport 31 (3), 343–358.
Acknowledgements The authors would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum and Minerals (KFUPM), Saudi Arabia for funding this work under Grant No. IN121065. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.cstp.2018.09.003. References Abate, M., Vierth, I., Karlsson, R., de Jong, G., Baak, J.A., 2018. Disaggregate stochastic freight transport model for Sweden. Transportation. https://doi.org/10.1007/ s11116-018-9856-9. Al-Atawi, A.M., 2005. Household Travel and Tour-based Behaviour: A Case Study from Saudi Arabia, Doctoral Dissertation, Cardiff University. Cai, P., Wang, Y., Lu, G., Chen, P., Ding, C., Sun, J., 2016. A spatiotemporal correlative knearest neighbor model for short-term traffic multistep forecasting. Transp. Res. Part C: Emerg. Technol. 62, 21–34. Chen, H., Grant-Muller, S., 2001. Use of sequential learning for short-term traffic flow forecasting. Transp. Res. Part C: Emerg. Technol. 9 (5), 319–336. Chen, X.Y., Pao, H.K., Lee, Y.J., 2014. Efficient traffic speed forecasting based on massive heterogenous historical data. In: 2014 IEEE Int. Conf. Big Data, pp. 10–17. Chou, J.S., Pham, A.D., 2013. Enhanced artificial intelligence for ensemble approach to predicting high performance concrete compressive strength. Constr. Build. Mater. 49, 554–563. Coombe, R.D., 1985. Urban transport policy development: two case studies in the Middle East. Transp. Rev. 5 (2), 165–188. Danech-Pajouh, M., Aron, M., 1991. ATHENA: a method for short-term inter-urban motorway traffic forecasting. Recherche Transports Sécurité 6, 6–11. De Jong, G., Daly, A., Pieters, M., Vellay, C., Bradley, M., Hofma, F., 2003. A model for time of day and mode choice using error components logit. Transp. Res. Part E: Logist Transp. Rev. 39 (3), 245–268. De Neufville, R., 2008. Low-cost airports for low-cost airlines: flexible design to manage the risks. Transp. Plan. Technol. 31 (1), 35–68. Dia, H., 2001. An object-oriented neural network approach to short-term traffic forecasting. Eur. J. Operat. Res. 131 (2), 253–261. Draper, P., Paudyal, K., 2002. Explaining Monday returns. J. Financ. Res. 25 (4), 507–520. Eisenhauer, J.G., 2003. Regression through the origin. Teach. Stat. 25 (3), 76–80. El-Alfy, E.S.M., Ratrout, N.T., Gazder, U., 2015. Impact of stock market indices and other regional exogenous factors on predictive modeling of border traffic with neural network models. Arab. J. Sci. Eng. 40 (2), 303–312. Feng, X., Saito, M., Liu, Y., 2016a. Improve urban passenger transport management by
6