Unsupervised Machine Learning for Managing Safety Accidents in Railway Stations

Abstract

Railway operations, both for passengers and freight, hinge on reliability, accessibility, maintenance, and safety (RAMS). Particularly in urban areas, the risk of safety accidents at railway stations poses a significant concern for daily operations. These accidents not only damage the reputation of the market but also result in injuries, anxiety among the public, and financial costs. With stations facing increased pressure due to higher demand, there's a greater strain on infrastructure and heightened considerations for safety administration. To address this challenge, it's proposed to leverage unsupervised topic modeling, specifically optimizing Latent Dirichlet Allocation (LDA), to better understand the factors contributing to severe accidents. By analyzing textual data from 1000 accidents at UK railway stations gathered by RSSB, this research aims to systematically identify accident characteristics using machine learning topic modeling techniques. The goal is to enhance safety and risk management by gaining insights into accident causes and identifying hotspots in stations. This approach allows for advanced analysis of accident history, lessons learned, and a deeper understanding of risk factors associated with fatalities. The study evaluates the effectiveness of this text mining approach in extracting valuable accident information, such as root causes and hotspots, with predictive accuracy. By leveraging big data analytics, it offers a comprehensive understanding of accident nature that goes beyond traditional narrow domain analysis of accident reports. This technology stands to Significantly improve safety in the railway industry and holds promise for applications in other safety-critical fields, ushering in a new era of AI-driven safety applications.