An Ensemble Deep Learning Model for Vehicular Engine Health Prediction

Abstract

Predictive maintenance is increasingly vital in the automotive sector to ensure engine reliability, safety, and cost 
efficiency. Modern engines generate complex, multivariate sensor data, making early detection of failures 
challenging. This study presents a comprehensive vehicular engine health monitoring framework using an 
ensemble of machine learning models, including Logistic Regression, Support Vector Machines, Random Forest, 
XGBoost, and deep learning approaches. The system incorporates rigorous preprocessing steps—duplicate 
removal, median-based imputation, feature scaling, and Synthetic Minority Over-sampling Technique (SMOTE) - to handle noisy and imbalanced datasets. Models were evaluated using metrics such as accuracy, precision, 
RMSE, MAE, confusion matrix, and AUC. Among all configurations, the Random Forest Classifier achieved the 
best performance with a balanced accuracy of 71%, demonstrating effective prediction of engine health under 
varying operating conditions. The framework was deployed via a Flask-based web application, enabling real
time predictions and early warnings of potential engine faults. This approach provides a scalable, practical 
solution for reducing maintenance costs and improving vehicular safety.