The Utilization of ML and XAI Methods for Anomaly Detection fro Event Logs in Automotive Industry using Python

The aim of this bachelor thesis is to provide a detailed overview of anomaly detection methods and to investigate their application to automotive industry data. The theoretical part discusses each method, including traditional statistical, distance and rule-based approaches, as well as modern machine learning (ML) approaches such as Isolation Forest and Local Outlier Factor (LOF). Furthermore, the importance of explaining black box solutions using Explainable Artificial Intelligence (XAI) methods is highlighted, allowing for better understanding and confidence in machine learning models. In the practical part of the thesis, these methods are applied to real data from the automotive industry concerning a part of the vehicle manufacturing process. For each method, a thorough evaluation of their performance, interpretation and suitability for specific anomaly detection tasks is performed. As part of this evaluation, different data preprocessing strategies are also investigated, including class imbalances that may affect the efficiency and reliability of anomaly detection model. Finally, the paper presents recommendations and limitations of different anomaly detection methods in the context of the automotive industry. These recommendations include optimal practices for deploying and tuning these methods, as well as possible directions for further research and development in anomaly detection and explainable artificial intelligence. The thesis contributes to current discussions and efforts to improve vehicle safety, reliability and performance and provides a useful basis for further research in this area.