Evaluation and Prioritization of Crash Hotspot Prediction Methods for Interurban Roads using an AHP-TOPSIS Approach

Document Type : Original Article

Authors
Shahin Shabani 1
Ali Ghorbani 2
Peyman Pejmanzad 3
1 Department of Civil Engineering, Payame Noor University, Tehran, Iran
2 Department of Civil Engineering‌, Payame Noor University (PNU), Tehran, Iran.
3 Ph.D., Student, Department of Civil Engineering‌, Payame Noor University (PNU), Tehran, Iran.
10.22034/road.2026.591556.2522
Abstract
Road accidents in Iran cause over 20,000 deaths annually and account for approximately 7% of GDP. Identifying crash hotspots is one of the most effective strategies to reduce these losses. Various methods from three categories—traditional statistical, machine learning, and deep learning—have been proposed for hotspot prediction. However, the lack of a systematic comparison makes it difficult to select the appropriate method for Iranian roads. Fifteen prediction methods, including 3 statistical (Crash Frequency, Crash Rate, Empirical Bayes), 4 machine learning (Random Forest, SVM, XGBoost, Decision Tree), and 8 deep learning (CNN, LSTM, CNN-LSTM, GCN, GAT, STGNN, Transformer, Ensemble), were evaluated across 7 criteria (accuracy, sensitivity, speed, data volume, interpretability, generalizability, technical expertise). Data were extracted from a systematic review of 90 articles (2019–2025). AHP was used for weighting and TOPSIS for ranking. The hybrid CNN-LSTM achieved the highest rank with a TOPSIS score of 0.852. XGBoost was the best machine learning method with 0.795. Statistical methods ranked last with a mean score of 0.275. The mean score for deep learning (0.652) was higher than machine learning (0.634) and statistical methods (0.275). Sensitivity analysis confirmed the stability of the results. The hybrid CNN-LSTM method was selected as the top-ranked method for predicting crash hotspots on Iranian interurban roads. In very limited data conditions (fewer than 3,000 records), Empirical Bayes is recommended, and in situations requiring high interpretability, Decision Tree is proposed.
Keywords
Black Hotspots
Deep Learning
Machine Learning
AHP-TOPSIS
-احسانی، سید مجید و آسوده، راضیه (۱۳۹۹). شناسایی نقاط حادثه خیز راه‌های برون شهری با استفاده از روش مدل‌سازی فازی. مجله راهور، ۴(۲)، 134-103.
-سادات حسینی، سید محمد؛ سلیمی کوچی، محمدباقر؛ جوزی دستجره، امیر و بیرانوند، محسن (۱۴۰۳). شناسایی و تحلیل نقاط حادثه خیز شهر همدان. مجله راهور، ۸(۱)، ۱۳۱-۱۵۲.
- غفاری، سارا و حسینعلی، فرهاد (۱۴۰۳). بررسی تأثیر عوامل هندسی بر شدت تصادفات در راه‌های برون‌شهری با استفاده از الگوریتم‌های هوش مصنوعی و یادگیری عمیق. پژوهشنامه حمل و نقل، ۲۱(۳)، ۴۵-۶۸.
-مرکز پژوهش‌های مجلس شورای اسلامی (۱۴۰۳). بسته سیاستی کاهش تصادفات رانندگی در سال ۱۴۰۳. تهران.
-سازمان پزشکی قانونی کشور (۱۴۰۲). گزارش آمار تلفات حوادث رانندگی کشور. تهران.
-AASHTO. (2010). Highway safety manual. AASHTO.
-Alagarsamy, S., et al. (2021). Prediction of road accidents using machine learning technique. IJARSCT, 6(1), 45-52.
-Anik, B. M. T. H., et al. (2023). inTformer:A time-embedded attention-based transformer for crash likelihood prediction. arXiv:2307.03854.
-Bradley, A. P. (1997). The use of the area under the ROC curve in the evaluation of machine learning algorithms. Pattern Recognition, 30(7), 1145-1159.
-Chang, L. Y., & Chen, W. C. (2005). Data mining of tree-based models to analyze freeway accident frequency. Journal of Safety Research, 36(4), 365-375.
-Chaudhary, M., et al. (2023). Risk maps and intervention strategies in road traffic accidents using GIS. Transport Safety Review, 52, 21-38.
-Elvik, R. (2008). A survey of operational definitions of hazardous road locations. Accident Analysis & Prevention, 40(6),1830-1835.
-Gao, Y., et al. (2023). STZITDGNN: Spatiotemporal zero-inflated Tweedie distribution graph neural network. Transportation Research Part C, 148, 104002.
-Golestani, N., Moosavi, S. M. H., & Shabani, S. (2025). Bus telematics data analysis for crash risk prediction on Iranian intercity roads. PLoS ONE, 20(4), e0312345.
-Hauer, E. (1996). Identification of sites with promise. Transportation Research Record, 1542, 54-60.
-Hauer, E. (1997). Observational before-after studies in road safety. Pergamon.
-Hwang, C. L., & Yoon, K. (1981). Multiple attribute decision making. Springer.
- Kaffash Charandabi, N., Gholami, A., & Abdollahzadeh Bina, A. (2022). Road accident risk prediction using generalized regression neural network optimized with self-organizing map. Neural Computing and Applications, 34(11), 8511-8524.
-Lipton, Z. C. (2018). The mythos of model interpretability. Queue, 16(3), 31-57.
-Lord, D., & Mannering, F. (2010).The statistical analysis of crash-frequency data. Transportation Research Part A, 44(5),291-305.
-Nippani, A., et al. (2023). Graph neural networks for road safety modeling. arXiv:2311.00164.
-Pan, Y., et al. (2018). Regularized deep belief network for traffic accident prediction. IEEE Transactions on Intelligent Transportation Systems, 19(12), 3890-3900.
-Rana, P., et al. (2024). Machine learning approach to enhance highway railroad grade crossing safety. Transportation Research Record, 2678(3), 245-260.
-Saaty, T. L. (1980). The analytic hierarchy process. McGraw-Hill.
-Sajadi, P., et al. (2024). Accident impact prediction based on deep convolutional and recurrent neural network. arXiv:2411.07537.
-Shabani, S., et al. (2019). A risk based model for the switch derailment using Bayesian network. Proceedings of the 4th International Conference on Railway Technology.
 -Shabani, S., et al. (2021). Developing a model for the risk of the rail vehicles collision using Bayesian network. Journal of Transportation Safety, 43(2), 145-162.
-Singh, G., et al. (2020). Deep neural network-based predictive modeling of road accidents. Procedia Computer Science, 167, 1672-1681.
-Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information Processing & Management, 45(4), 427-437.
-Theofilatos, A., et al. (2019). Comparing machine learning and deep learning methods for real-time crash prediction. Accident Analysis & Prevention, 132, 105278.
-Wan, R., et al. (2023). Traffic accident hotspot prediction using temporal convolutional networks. In ACM SIGSPATIAL 2023.
-Yang, S., & Abdel-Aty, M. (2025). Real-time crash prediction with limited connected vehicle data using ensemble LSTM-CNN. Transportation Research Part C, 162, 104112.
-Yu, L., et al. (2021). Deep spatiotemporal graph convolutional network for traffic accident prediction. Neurocomputing, 423, 135-147.
-Yuan, J., et al. (2019). Real-time crash risk prediction using LSTM recurrent neural network. Transportation Research Record, 2673(4), 314-326.