Ensemble Gradient Boosting Classifier for Effective Identification of Software Faults

Timely identification of fault-prone modules is essential for improving software quality, enhancing customer satisfaction, and optimizing testing efforts. This study explores fault prediction, focusing on the development of robust classification models using established software metrics. In machine learning, ensemble classifiers are known to improve the accuracy of individual models by combining multiple learners. A comparative analysis is performed between ensemble learning methods—bagging and stacking—and the gradient boosting classifier, leveraging the widely used JM1 dataset from the NASA PROMISE repository. The evaluation, based on F1-score and ROC-AUC metrics, offers practical guidance for software teams in selecting effective models for fault detection. By promoting the use of sophisticated ensemble techniques, this research supports the creation of more dependable and maintainable software systems.