Predicting Breakdowns in Transportation Vehicles using Supervised Learning

Sammanfattning: Vehicle breakdowns can lead to fatal accidents, increase costs and reduce productivity. Therefore, robust and accurate fault diagnosis and prediction systems are critical to ensure the proper operation of vehicles. Many researchers have used machine learning for the prediction of vehicle breakdowns. However, previous work has typically predicted the failure of a specific component or predicted whether a breakdown will occur without predicting the specific type of breakdown. This thesis used ensemble models to predict breakdowns that have not yet occurred through vehicle breakdown data. The model proposed in the thesis is able to predict six common types of vehicle breakdowns. This study faces the problem of data imbalance that may cause some of the breakdowns difficult to predict. This work used two methods to balance the classes and various ensemble models to solve this problem and compares the results with baseline models such as Support Vector Machine (SVM) and Random Forest Classifier (RF) to demonstrate the powerful predictive power of ensemble models.