Performance Comparison of Ensemble and Supervised Models for Software Defect Prediction

Software defect prediction (SDP) is one of the significant activities for the quality and reliability of software systems. Ensemble learning techniques have recently become very popular because they combine the strength of more than one base Machine Learning model to achieve improved accuracy. In this study, we compared an ensemble-based approach for SDP with individual base models. We applied a robust ensemble model by combining Decision Trees, Random Forests, Gradient Boosting, and Support Vector Machines. Then, we evaluate the performance using 5 public datasets from NASA's Promise repository and compare the results against the different individual models. Results validate that our ensemble approach has higher accuracy (up to 94.60%), precision, recall, and F1 score than most of the standalone models. We will give insights on the superior performance of the Ensemble model for SDP tasks. Our results underline the potential of ensemble learning in improving software defect prediction.