Predicting the Segment-Based Effects of Heterogeneous Traffic and Road Geometric Features on Fatal Accidents

Inter-urban roads in Indonesia are characterized mainly by distinct road geometry and heterogeneous traffic features. The accident database from the Republic of Indonesia National Traffic Police recorded a substantial number of fatal accidents and fatalities along inter-urban roads. This study aimed to analyze the effects of traffic heterogeneity and road geometry features on fatal accidents along inter-urban roads in South Sulawesi, Indonesia. Segment-based accident analysis was adopted to minimize bias due to the large standard deviations of road lengths. Vehicle-specific speeds, speed standard deviations, and volumes of six vehicle categories, road surface condition, and road geometry were the classified predicting factors. A machine learning technique was adopted to produce predictions of the classification problem. A total of 1,068 road segment observations from 2013–2016 were used to build and validate the model. Model generalization was carried out using the out-of-sample 2019 data. With 26 potential predictors, three machine learning techniques based on the ensembles of regression trees were used to avoid removing potential predictors altogether. The results indicate that road-related features show the greatest importance in predicting the number of fatal accidents. Among the speed features, the average speed of angkots and speed standard deviation of motorcycles showed the greatest importance. The average daily traffic (ADT) of pickups had the greatest importance among other vehicle-specific ADTs.

Fatal accidents; Heterogeneous traffic; Machine learning; Segment-based effects; Speed standard deviation

Traffic accidents on inter-urban roads in Indonesia are still considered a serious problem, with high rates of fatalities. As an illustration, the fatality rates on Bantaeng–Bulukumba, Jeneponto–Bantaeng, and Bulukumba–Tondong in South Sulawesi were 23.6, 6.5, and 5.1 deaths per 100 million vehicle-km in 2015 (Australia–Indonesia Partnership, 2017) and were 7.03, 3.35, and 8.19 deaths per 100 million vehicle-km in 2019, based on the present study.

Various factors related to traffic safety have been studied using different approaches. Studies on speed variations have revealed that both the standard deviation (SD) of speed and the coefficient of speed variation (CSV) were significantly related to traffic collisions (Choudhary et al., 2018; Wang et al., 2018). The effects of road geometry on traffic safety were studied by Chen et al. (2019) and Papadimitriou et al. (2019), and the effects on pedestrian safety were studied by Siregar et al. (2019). In these studies, the traffic factors were considered homogeneous, and vehicle-specific traffic speeds and volumes were not considered. The heterogeneity of traffic in speeds and volumes was adopted in the study on fatality rates and accident rates (Siregar et al., 2020) on inter-urban roads. Similarly, various approaches have been applied to analyze traffic safety indices. Some of the most widely used regression analyses include the Poisson regression (Mitra et al., 2017) and negative binomial regression (Poisson-gamma; Tjahjono, 2010; Gitelman et al., 2017). An artificial neural network was also used in a study by Alkheder et al. (2017).    

    Because the distribution of accidents is substantially zero inflated and most road segments under study (86% in 2013–2016 data) did not have fatal accidents, the distribution is instrumental in predicting fatal accidents on a segment base that can be applied to fatal accident preventive measures. Therefore, the present study aimed to analyze the segment-based effects of traffic and road factors on predicting the number of fatal accidents on inter-urban roads by considering the heterogeneity of traffic. These factors, referred to as “features” in machine learning, include the average daily traffic, average speeds, and speed standard deviations of the different categories of vehicles, road geometry, road surface condition, and road section length. Machine learning was adopted for the analysis, given that the technique can overcome the nonlinear relationship problem between fatal accidents and predictive features. Combinations of the different features that are hypothesized to determine the number of accidents were explored. The findings can be expected to contribute to the development of traffic safety improvement programs. 

    The results indicate that there are different factors determining the number of fatal accidents. ADT, speeds, and speed SDs are vehicle-specific variables with different levels of importance. The findings of the present study should also be intuitively interpreted, considering that there might be some underreporting of fatal accidents due to failure to change the status from severe to fatal. Because the high contributions of some vehicle variables in an accident do not necessarily indicate the involvement of the vehicles in the accidents, further analysis is required to give a more integrated and comprehensive view of the relationships between different categories of vehicles and accidents. Having the greatest importance in percentage among the speed features, the average speed of angkots and the ADT of passenger cars and motorcycles necessitate special treatment to reduce the effects on fatal accidents. The results of this study can help traffic safety agencies target specific features for improvement measures.

    This research was funded by TADOK Grant Universitas Indonesia 2019 No. NKB-0167/UN2.R3.1/HKP.05.00/2019. The authors would also like to thank the Project Management Unit of the Australian-funded EINRIP Monitoring & Evaluation Programme, Fifth Monitoring Survey, Final Report 2017 for permission to use the data.

Alkheder, S., Taamneh, M., Taamneh, S., 2017. Severity Prediction of Traffic Accident using an Artificial Neural Network. Journal of Forecasting, Volume 36(1), pp. 100–108

Australia–Indonesia Partnership, 2007. EINRIP Alignment Layout

Australia–Indonesia Partnership, 2017. EINRIP Monitoring & Evaluation Programme, Fifth Monitoring Survey, Final Report

Chen, S., Saeed, T.U., Alinizzi, M., Lavrenz, S., Labi, S., 2019. Safety Sensitivity to Roadway Characteristics: A Comparison across Highway Classes. Accident Analysis & Prevention, Volume 123, pp. 39–50

Choudhary, P., Imprialou, M., Velaga, N.R., Choudhary, A., 2018. Impacts of Speed Variations on Freeway Crashes by Severity and Vehicle Type. Accident Analysis & Prevention, Volume 121, pp. 213–222

da Costa, D.G.N., Malkhamah, S., Suparma, L.B., 2018. Use of the Safety Factor and Margin of Safety in Motorcyclist Accident Risk Management. International Journal of Technology, Volume 9(4), pp. 737–750

Gitelman, V., Doveh, E., Bekhor, S., 2017. The Relationship between Free-Flow Travel Speeds, Infrastructure Characteristics and Accidents, on Single-Carriageway Road. Transportation Research Procedia, Volume 25, pp. 2026–2043

IRSMS Korlantas POLRI., n.d. Available Online at https://irsms.korlantas.polri.go.id/login. Accessed on March 23, 2019

Krishnaveni, S., Hemalatha, M., 2011. A Perspective Analysis of Traffic Accident Using Data Mining Techniques. International Journal of Computer Applications, Volume 23(7), pp. 40–48

Lee, J., Yoon, T., Kwon, S., Lee, J., 2020. Model Evaluation for Forecasting Traffic Accident Severity in Rainy Seasons using Machine Learning Algorithms: Seoul City Study. Applied Sciences, Volume 10(1), pp. 1–15

Mitra, S., Haque, M., King, M.J., 2017. Effects of Access, Geometric Design, and Heterogeneous Traffic on Safety Performance of Divided Multilane Highways in India. Journal of Transportation Safety & Security, Volume 9(Sup1), pp. 216–235

Molnar, C., 2020, Interpretable Machine Learning: A Guide for Making Black Box Models Interpretable. Leanpub

Papadimitriou, E., Filtness, A., Theofilatos, A., Ziakopoulos, A., Quigley, C., Yannis, G., 2019. Review and Ranking of Crash Risk Factors Related to the Road Infrastructure. Accident Analysis & Prevention, Volume 125, pp. 85–97

Pradhan, B., Sameen, M.I., 2020. Predicting Injury Severity of Road Traffic Accidents using a Hybrid Extreme Gradient Boosting and Deep Neural Network Approach. In: Laser Scanning Systems in Highway and Safety Assessment, Springer International Publishing, Cham, Switzerland, pp. 119–127

Siregar, M.L., Alawiyah, T., Tjahjono, T., 2015. Remedial Safety Treatment of Accident-Prone Locations. International Journal of Technology, Volume 6(4), pp. 680–688

Siregar, M.L., Sumabrata, R.J., Kusuma, A., Samosir, O.B., Rudrokasworo, S.N., 2019. Analyzing Driving Environment Factors in Pedestrian Crashes Injury Levels in Jakarta and the Surrounding Cities. Journal of Applied Engineering Science, Volume 17(4), pp. 482–489

Siregar, M.L., Tjahjono, T., Nahry, 2020. Endogenous Relationship of Accident Occurrence with Speed, Traffic Heterogeneity and Driving Environment on Inter-Urban Roads in Indonesia. Journal of Applied Engineering Science, Volume 18 (4), pp. 475–484

Sutton, C.D., 2005, Classification and Regression Trees, Bagging, and Boosting. In: Handbook of Statistics Volume 24, Rao, C.R., Wegman, E.J., Solka, J.L. (eds.), Amsterdam, The Netherlands, Elsevier, pp. 303–329

Tjahjono, T., 2010. The Effect of Traffic and Road Conditions to the Fatality Rates on Rural Roads in Eastern Indonesia. Journal of the Eastern Asia Society for Transportation Studies, Volume 8, pp. 2201–2213

Wang, X., Zhou, Q., Quddus, M., Fan, T., Fang, S., 2018. Speed, Speed Variation and Crash Relationships for Urban Arterials. Accident Analysis & Prevention, Volume 113, pp. 236–243

Zhang, J., Li, Z., Pu, Z., Xu, C., 2018. Comparing Prediction Performance for Crash Injury Severity Among Various Machine Learning and Statistical Methods. IEEE Access, Volume 6, pp. 60079–60087

Zuraida, R., Iridiastadi, H., Sutalaksana, I.Z., 2017, Indonesian Drivers’ Characteristics Associated with Road Accidents. International Journal of Technology, Volume 8(2), pp. 311–319