20.500.12556/DKUM-97433 A comparative analysis of traffic accident frequency models for motorway tunnels using machine learning Motorway tunnels are critical elements of the motorway network where traffic accidents, although less frequent than on open sections, often have more severe consequences. This paper develops a model for expected accident frequency in motorway tunnels based on a traditional Negative Binomial (NB) regression model and two machine-learning–based models: Neural Networks (NN) and Random Forest (RF). The study uses historical accident and traffic data for all motorway tunnels between 2013 and 2024, combined with key infrastructural characteristics. The analysis considers all recorded traffic accidents in motorway tunnels, including accidents with material damage only as well as injury-related accidents of varying severity. A stepwise procedure was used to determine the optimal NB model, resulting in a final specification with tunnel length and Annual Average Daily Traffic (AADT) as predictors. Machine-learning–based models were trained on the same input set and evaluated against the NB model using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and log-likelihood. The NN model achieved the lowest MAE (1.03 accidents/year), followed by RF (1.07) and NB (1.13), confirming that machine-learning-based (ML-based) models slightly improved predictive accuracy while maintaining interpretability at the network level. Compared to the NB model, the NN model achieved a reduction in Mean Absolute Error of 9.2%, while the RF model achieved a reduction of 5.1%. A detailed case study of the Trojane tunnel demonstrates that all models reproduce long-term accident trends, while machine-learning-based models better capture variations between years. The proposed modelling framework provides a practical decision–support tool for tunnel operators and policy makers by supporting tunnel risk classification, prioritization of safety investments, and medium-term safety planning. By supporting proactive tunnel safety planning and evidence-based prioritization of safety investments, the proposed framework contributes to sustainable transport infrastructure management. Improved prediction of accident frequency enables more efficient allocation of resources, reduction in accident-related social and economic costs, and enhanced long-term resilience of motorway tunnel systems. motorway tunnels machine-learning-based models Negative Binomial model accident prediction sustainable transport infrastructure sustainability avtoceste cestni predori modeli strojnega učenja napovedovanje nesreč trajnostni prevoz trajnostni transport odpornost infrastrukture trajnostna mobilnost true false true MDPI Angleški jezik Slovenski jezik https://www.mdpi.com/journal/sustainability © 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. Članek v reviji 2026-03-06 09:25:34 2026-03-06 09:25:35 2026-03-07 03:08:52 0000-00-00 00:00:00 2026 0 0 str. 1-21 št. 5, št. članka 2223 Let. 18 Feb. 2026 0000-00-00 Zaloznikova Objavljeno NiDoloceno 2026-01-14 2026-02-23 2026-02-25 656.11:004.85 270706435 10.3390/su18052223 2071-1050 sustainability-18-02223-v2.pdf sustainability-18-02223-v2.pdf 1 E3F78662B654C0B70210510BF12E90EC 544062173bd052523a3b103359668c340cd113ea62cee244ff57c88b8922fa55 82584fb3-1936-11f1-96ca-00155d580e1a https://dk.um.si/Dokument.php?lang=slv&id=204323 https://www.mdpi.com/2071-1050/18/5/2223 1 0e523b80-1936-11f1-96ca-00155d580e1a https://dk.um.si/Dokument.php?lang=slv&id=204322 Fakulteta za gradbeništvo, prometno inženirstvo in arhitekturo 0 0 0