Ensemble-based Machine Learning Approach for Automated Software Defect Prediction

Abstract

In the tech industry, ensuring software reliability is a critical concern for professionals, often addressed through traditional techniques that rely on prior experience or identifying faulty modules within an application. These methods can be time-consuming and may not always pre-emptively address issues. Automated software defect prediction models, driven by ensemble learning techniques, offer a proactive approach to significantly enhance a software's ability to predict and mitigate defects, leading to more efficient operation, reduced errors, and lower costs. This paper proposes a software defect prediction model based on ensemble learning methods, aimed at maintaining software functionality more effectively. Using established evaluation benchmarks including ten-fold cross-validation, precision, recall, specificity, F1 measure, and accuracy our study evaluates the performance of various machine learning algorithms: Ensemble Learning (EL), Decision Trees (DT), Naive Bayes (NB), Artificial Neural Networks (ANN), and Support Vector Machines (SVM). The results reveal that EL consistently outperforms other models with classification accuracy ranging from 98% to 100%, demonstrating its robustness and superior ability to balance precision and recall across diverse datasets (JM1, CM1, and PC1). Following EL, DT also performs strongly but with slightly lower accuracy, particularly in contexts where interpretability is crucial. NB and ANN show decent results but require careful tuning to achieve optimal performance, while SVM ranks lowest in this analysis. These findings underscore the importance of selecting and implementing appropriate algorithms based on the specific demands of software defect prediction tasks, with EL emerging as the most reliable and robust choice for enhancing software reliability.