Identifying climate-related failures in railway infrastructure using machine learning

Climate change impacts pose challenges to a dependable operation of railway infrastructure assets, thus necessitating understanding and mitigating its effects. This study proposes a machine learning framework to distinguish between climatic and non-climatic failures in railway infrastructure. The maintenance data of turnout assets from Sweden's railway were collected and integrated with asset design, geographical and meteorological parameters. Various machine learning algorithms were employed to classify failures across multiple time horizons. The Random Forest model demonstrated a high accuracy of 0.827 and stable F1-scores across all time horizons. The study identified minimum-temperature and quantity of snow and rain prior to the event as the most influential factors. The 24-hour time horizon prior to failure emerged as the most effective time window for the classification. The practical implications and applications include enhancement of maintenance and renewal process, supporting more effective resource allocation, and implementing climate adaptation measures towards resilience railway infrastructure management.