Adaptive Intelligence in Fault Prediction: Tackling Imbalance and Generalization Challenges via Cross- Project Learning

Software fault prediction (SFP) is a critical exercise to guarantee the reliability of software as well as lower the maintenance expenses. Nonetheless, the key causes of the current fault prediction model being as cumbersome with adopting to other projects as they are are, class imbalance and low generalisation ability. The system introduced in this paper is a system with adaptive intelligence that resides on a hybrid resampling framework/transfer learning framework that is created to solve the issue of imbalance in data and enhance fault prediction model completeness. The proposed model will use ensemble based meta-learners to choose dynamically, weight classifiers based on measures of similarity between inter-projects in order to help increase the generalisation perspective of the models in relation to other software projects. It was experimentally tested on a used-set open-source data of PROMISE and NASA show that the proposed model was superior to the traditional within-project models on the F1- score, recall and AUC performance metrics. The study results indicate that the adaptive intelligence strategies enhance the transferability of the model, as well as mitigate the problem of class imbalance to better the automation and scalability of software engineering practice prediction of failover.