Anomaly detection on vehicle data: models and explainability

Vehicle transportation is essential in today’s society. It facilitates the movement of people and goods at a reasonable cost and with high efficiency, particularly for short to medium distances. In this context, managing and preventing vehicle failures becomes crucial for key performance indicators (KPIs) such as safety, efficiency, and the reliability of modern transportation systems. As vehicles become increasingly sophisticated with advanced sensors and complex onboard systems, data collection tools play a crucial role in preventing and detecting failures to avoid subsequent system breakdowns that could impact the proper functioning of vehicles. In this thesis, we will introduce a new tool designed to provide companies with insightful analysis of specific failures. This tool involves the development of a machine-learning model that can be applied directly to new vehicle data to identify potential anomalies, thereby preventing further failures. Additionally, it aims to offer an understanding of how the model utilizes input features to predict output results through explainable AI frameworks. This approach spans from predicting engine malfunctions before they occur to detecting unusual driver behavior in real time and understanding the potential factors contributing to these issues. Ultimately, this research aims to contribute to safer roads, more efficient transportation systems, and a more sustainable future. This thesis seeks to make a valuable contribution to the field of vehicle transportation by introducing a new tool designed to provide companies with insightful analysis of specific failures.