Detecting and Managing Railway Station Safety Accidents Using Unsupervised Machine Learning

Ensuring the reliability, accessibility, maintainability, and safety (RAMS) of railway stations is vital for both passenger and freight operations. As urbanization and transit demands increase, railway stations face heightened safety risks and operational challenges. Accidents in these environments not only cause injuries and fatalities but also result in reputational damage and financial costs. This study explores the application of unsupervised machine learning— specifically Latent Dirichlet Allocation (LDA) topic modeling—to analyze unstructured textual accident reports sourced from the UK Rail Safety and Standards Board (RSSB), comprising data from 1,000 recorded station-related incidents. By extracting hidden patterns and identifying recurring themes in fatality-related accidents, this approach offers a data-driven method to uncover root causes and high-risk areas within stations. The analysis supports predictive insights that can enhance risk assessment and improve proactive safety management. Leveraging intelligent text mining techniques allows for a broader, more comprehensive understanding of safety issues than traditional case-by-case reviews. This work contributes to advancing AI-based solutions in transportation safety, offering scalable, accurate insights that support strategic planning and real-time decision-making in the railway sector.