Data-Driven Bayesian Network Analysis of Railway Accident Risk

Ensuring railway safety is a top priority, with a central focus on preventing accidents. By thoroughly analyzing data from railway accident investigations, we can pinpoint factors and patterns associated with different types of railway accidents. This proactive approach not only helps reduce the frequency of such incidents but also significantly boosts overall railway transportation safety. This paper investigates the impact of various risk factors on railway safety through the analysis of railway accidents by using data-driven Bayesian networks. First, key data representing the frequency of risk factors directly derived from railway accident reports are collected and analyzed. Then, the risk factors are incorporated into causal analysis for different types of railway accidents. Finally, a historical data-driven approach is utilized to model and gain new insights into the key risk factors causing different types of railway accidents. Meanwhile, a Tree-Augmented Naive Bayes (TAN) is employed to construct a model of interdependencies among risk factors, and the model is validated through sensitivity analysis and past accident records. The research findings demonstrate that the crucial risk factors for all types of accidents include undetected track damage, train operator skills, load, braking system conditions, train speed, traction system failures, level crossings, and bridge damage. Additionally, the research results highlight the differential impact of key factors on different types of accidents, providing a most probable explanation for observing the most likely configurations in the model for a specific scenario. This work contributes to accident prevention and safety decision-making.