CROSS-PROJECT SOFTWARE FAULT PREDICTION ADDRESSING CLASS IMBALANCE AND GENERALIZATION

Abstract

The software development industry is notoriously challenging due to the unpredictability of faults and the substantial costs associated with rectifying them post-implementation. CPSFP is a novel methodology for software fault prediction that leverages data from prior projects to forecast the occurrence of faults in a current project. There exists a pervasive absence of order. Although software issues are significantly rarer than evident cases, a primary concern is that predictive algorithms are ineffective for learning. The great variety of tasks indicates that talents developed in one domain may not be transferable to another. This research utilizes domain adaptation techniques, robust machine learning models, and advanced dataset balancing algorithms such as SMOTE to address these challenges. The aim is to enhance the precision of predictions and broaden their applicability across various project types. Our methodology improved mistake detection and generalization when utilized across several open-source datasets. Software quality assurance may be improved if teams were able to spot issues sooner.