MACHINE LEARNING BASED HYBRID APPROACH FOR SOFTWARE DEFECT PREDICTION

Ensuring software reliability is a critical challenge in the tech industry, traditionally addressed through manual inspection and experience-based techniques that can be time-consuming and inefficient. Automated software defect prediction models, leveraging machine learning, offer a proactive solution to identify and mitigate defects early in the development cycle. This study proposes a defect prediction model based on hybrid learning techniques, evaluating its performance against Decision Trees (DT), Naïve Bayes (NB), Artificial Neural Networks (ANN), and Support Vector Machines (SVM). Using standard evaluation metrics such as ten-fold cross-validation, precision, recall, specificity, F1-score, and accuracy, our findings demonstrate that hybrid learning consistently outperforms other models, achieving classification accuracy between 98% and 100% across multiple datasets (JM1, CM1, and PC1). While DT also performs well, NB and ANN require careful tuning, and SVM exhibits the lowest accuracy. These results highlight hybrid learning as a robust and effective approach for enhancing software reliability by improving defect prediction accuracy.