Unsupervised Machine Learning For Monitoring Unsafe Events In Railway Zones

Abstract

Railway station safety is a critical aspect of transportation operations, especially in densely populated urban areas where increasing demand puts additional pressure on infrastructure. Accidents at stations can result in fatalities, injuries, public anxiety, reputational damage, and financial loss. This research presents an AI-based approach using unsupervised machine learning, specifically Latent Dirichlet Allocation (LDA), to analyse and understand the underlying factors contributing to fatal accidents in railway stations. A dataset of 1,000 fatal accident reports from Indian railway stations, sourced from the RSSB, is used for analysis.
The study aims to identify hidden patterns, root causes, and high-risk zones through topic modelling, offering a systematic way to enhance risk assessment and safety management. By leveraging intelligent text mining, the research extracts valuable insights from historical data, providing predictive accuracy and supporting data-driven decision-making. The findings demonstrate how AI and big data analytics can improve railway safety, moving beyond traditional, narrow analysis methods and ushering in a new era of safety intelligence in the transportation sector.