A Principal Component Analysis (PCA)-Deep Neural Network Model for Automated Anomaly Detection Using Metro Turnout Data

Abstract— Transportation engineering is being transformed with the advent of data driven statistical models. Additionally, industry 4.0 has changed anomaly detection and predictive maintenance using data from sensors and monitoring systems. Urban metro systems form the backbone of public transportation in many cities, ensuring timely and safe movement of millions of passengers. Among the key infrastructural components of a metro network are turnouts, also known as railway switches, which allow trains to move from one track to another. The performance and reliability of turnouts directly affect the efficiency and safety of the rail system. However, due to their mechanical and electrical complexity, turnouts are prone to failures and malfunctions. Automated anomaly detection using metro turnout data has emerged as a critical approach to address these challenges, enabling proactive monitoring and predictive maintenance. This paper presents a combination of Principal Component Analysis (PCA) and Deep Neural Network based statistical model for automated anomaly detection using metro turnout data. It has been shown that the proposed work attains lower error percentage compared to existing research frameworks in the domain. Keywords— Transportation Engineering, Statistical Modelling, Metro Turnout Imbalanced Datasets, Principal Component Analysis (PCA), Deep Neural Network, Error Percentage.