Predicting Electric Vehicle Failures Using Machine Learning: A Comparative Study on Imbalanced Sensor Data

Santa Cruz Crespo, Alexis Reyna (2025) Predicting Electric Vehicle Failures Using Machine Learning: A Comparative Study on Imbalanced Sensor Data. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (692kB) | Preview

Abstract

The use of Electric vehicles (EVs) is increasing but some unexpected technical failures can affect safety, reliability and operating costs. Traditional maintenance methods are often not able to detect early failures and it creates the need for predictive maintenance strategies. There are some limitations for machine learning models in EVs when they are working with imbalanced data because it can affect the performance. This study evaluates and compares four machine learning models: Logistic Regression, Decision Tree, Random Forest and XGBoost to predict Diagnostic Trouble Codes (DTC) using EV sensor data.

The project also studies the use of Synthetic Minority Oversampling Technique (SMOTE) to improve the performance of the model. The dataset has 131,396 units from three different profiles of EV users (rare, moderate and heavy), the dataset was cleaned, merged and processed. The models were trained using both datasets, imbalanced and balanced and they were evaluated using different evaluation metrics such as Accuracy, Balanced Accuracy, Precision, Recall, F1 Score, and ROC AUC. The results showed that without balancing the data, Logistic Regression could not detect failures and the tree-based models got high accuracy with a possible overfitting. After applying SMOTE, all the models improved their performance. Random Forest and XGBoost showed the best results between precision and recall and these results confirm that addressing class imbalance is very important for predictive failures in EVs and support the use of machine learning models for predictive maintenance to reduce downtime and costs. In the future we can study deep learning models with real-time data.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Onwuegbuche, Faithful Chiagoziem UNSPECIFIED
Uncontrolled Keywords:	Electric Vehicles; Predictive Maintenance; Machine Learning; SMOTE
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Artificial Intelligence for Business
Depositing User:	Ciara O'Brien
Date Deposited:	24 Jun 2026 11:41
Last Modified:	24 Jun 2026 11:41
URI:	https://norma.ncirl.ie/id/eprint/9404

Actions (login required)

View Item