A COMPARATIVE STUDY ON SOFTWARE DEFECT PREDICTION USING ML TECHNIQUES

Abstract

Software bug forecast is a key factor in software engineering which focuses on detecting modules that may have defects before any further development. More precise predictions about software bugs can raise the product quality, make it more reliable and cheaper to maintain. At this juncture, I would like to give an exhaustive analysis of different machine learning methods available regarding software defect prediction. This research work mainly focuses on investigating how well some of these techniques fare when applied on datasets obtained from PROMISE which is an online repository that has several standard datasets commonly used by researchers in the corresponding field. Also included in the analysis are algorithms such as Decision Trees, Support Vector Machines, Neural Networks or Random Forests among others; all these however shall be based on k-nearest neighbors (KNN). The procedure involves a thorough process of collecting information, pre-processing it, and choosing features to guarantee that datasets are prepared well for training and evaluating models effectively. On top of that, we utilize strict cross-validation techniques for assessing the performance of the built models by making sure their validity and reliability (are acceptable).Using different machine learning methods, the performance metrics such as accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), and Area Under the ROC Curve (AUC) are used for accessing the results. The findings point to ensemble strategies being more efficient in terms of predictive accuracy and generalizability than individual classifiers across different runs, especially Random Forests and Gradient Boosting Machines. By conducting a thorough comparative analysis of machine learning methodologies for software bug prediction, this research contributes to the software engineering discipline. The results show that not only do ensemble methods have the most impressive results, but we should also take into consideration such issues as interpretability, computational resources, and characteristics of the particular software project while choosing a method to use.