Predictive maintenance optimisation for CCTV systems in electric multiple unit trains using machine learning techniques

This paper presents the study conducted to investigate the maintenance strategy to improve the reliability of Closed-Circuit Television (CCTV) systems in railway rolling stock Electric Multiple Unit trains. The project attempts to optimise maintenance procedures by assessing 1214 sample datasets collected from sensors and control units during fleet data processing to identify and forecast CCTV faults. The analysis indicates that the CCTV system is the worst-performing system leading to delay and cancellation of service operations. The study attempts to address the pattern of CCTV failures using predictive modelling tools, and machine learning techniques such as Random Forest Regressor, Gradient Boosting Regressor, XGBoost Regressor, and Decision Tree Regressor used for modelling and prediction. The results exhibit satisfactory predictive accuracy of the incident reported days, starting date, and issue date for each incident, the results show important performance indicators such as Mean Squared Error, R-squared and Mean Absolute Error, which indicate promising outcomes. The results emphasise the capability of predictive modelling to improve the dependability of CCTV systems in railway rolling equipment, leading to enhanced operational efficiency and passenger safety.