Managing Safety Accidents in Railway Stations using Topic Modeling and Predictive Analysis

Abstract Railway stations are high-density areas where safety is a prime concern due to frequent operational risks, infrastructure limitations, and increasing passenger load. This paper proposes a machine learning-based system using Natural Language Processing (NLP) and topic modeling to analyze and predict safety incidents. Accident reports from the UK RSSB dataset are preprocessed and analyzed using Latent Dirichlet Allocation (LDA) to extract hidden topics and patterns. Additionally, classifiers such as CatBoost and Voting Classifier are trained on the transformed data to predict accident types. Our proposed system enables risk identification, real-time analysis, and visualization for railway safety management. Results show high accuracy and demonstrate the potential of AI-enhanced decision support in public safety infrastructure. Keywords: Railway Safety, Accident Prediction, Topic Modeling, LDA, NLP, CatBoost, Machine Learning