Explainable Machine Learning Techniques for Predictive Car Engine Health Monitoring

^1
1
Makerere University, College Of Computing and Information Sciences Research, Innovations, Services, and Engagements (RISE) program.Modern automobiles' increasing complexity demands real-time engine diagnostics to ensure optimal performance and safety. This research proposes a solution utilizing machine learning models trained on historical data to detect potential engine faults and anomalies. The goal is to facilitate preventive maintenance, minimize unscheduled breakdowns, and enhance overall vehicle reliability. This research study explores various machine learning models, including Support Vector Machines (SVM), Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), and ensemble models. Sensor data serves as the primary input, alongside various evaluation metrics like accuracy, precision, recall, and Fl-score. Ensemble models explored include combinations of ANN, SVM, linear regression, CatBoost, XGBoost, LightGBM, and AdaBoost with Gradient Boosting. Explainability techniques were investigated, with LIME chosen for its ease of use and interpretability. Amongst the evaluated models, KNN achieved the highest performance with an accuracy of 0.6318, precision of 0.5660, recall of 0.4770, and Fl-score of 0.4678. However, current techniques face challenges like noisy sensor data and incompatibility with existing engine systems. This research study aims to overcome these limitations and significantly improve predictive monitoring, vehicle reliability, and real-time engine health diagnosis.