Applying probabilistic mathematical modeling approach and AI technique to investigate serious train accidents in Japan

Abstract

We investigated data for serious train accidents (STAs) in Japan to elucidate their causes and consequences and to improve countermeasures for reducing the number of STAs. We used statistical data on the STAs occurring in Japan from 1987 to 2018, which included the frequency, types, causes, and consequences of the STAs, along with additional derailment, collision, and casualty data. We investigated the historical trend of the STAs using various probabilistic mathematical modeling approaches, such as Markov models, logit regression models, Bayesian approaches, and artificial-intelligence techniques. We showed that the number of casualties in STAs involving collisions was significantly larger than that for accidents not involving collisions. Thus, the statistical analysis indicated that preventing train collisions is the most important and necessary measure for reducing damage to passengers. Additionally, we proposed several countermeasures for ensuring the safety of passengers in Japan, e.g., install automatic train stops for all railway companies of Private Railway and terminate the use of ground-level crossings without gates. We evaluated the effectiveness of these countermeasures from various viewpoints.