A data-driven framework for railway risk assessment and safety management: evidence from Thailand’s national network

This study presents a data-driven framework to assess risk and enhance safety within Thailand’s railway network, particularly during Thailand’s transition to high-speed rail. Using more than 70,000 accident records (2009–2024), the research applies temporal, spatial, and statistical analyses to identify high-risk patterns. Findings indicate that accidents peak during the evening hours and the rainy season, with derailments and level-crossing collisions significantly associated with higher fatality rates (r = 0.68, p < 0.01). At the provincial level, Bangkok recorded the highest number of incidents (n = 31) but comparatively few casualties, while Phetchaburi and Chachoengsao reported fewer incidents yet disproportionately high numbers of injuries and fatalities. To address these risks, the study employs Decision Tree, Random Forest, and Bayesian Network models. The Random Forest achieved the highest predictive accuracy (95.6%), while the Bayesian Network offered interpretable causal reasoning. These models underpin the Rail-Risk Management Platform, a web-based tool providing real-time visualisation, alerts, and scenario analysis. Pilot testing reduced assessment time from 45 to 8 minutes and achieved a user satisfaction score of 4.52 out of 5. The study recommends targeted investments, intelligent safety systems, workforce development, and a unified risk management framework for sustainable railway-safety governance in Thailand.