Efficient Machine Learning Approach Based Bug Prediction for Enhancing Reliability of Software and Estimation

Abstract

Software Bug Prediction (SBP) serves as an essential process that predicts potential problems before software deployment to achieve software success. The detection of issues during development dramatically advances system performance quality and reduces needs for financial investments. The accuracy of software bug prediction has been greatly enhanced and expenses have been decreased with the inclusion of Machine Learning (ML) methods. A technique based on Deep Neural Networks (DNNs) is suggested in this research to enhance the accuracy of defect prediction. The method involves preprocessing the CM1 NASA dataset followed by ANOVA F-test feature selection and resolving class imbalance through ADASYN implementation. The DNN model gained 95% accuracy during training with optimized hyperparameters and exceeded both traditional Support Vector Machines with 87% accuracy and Random Forest with 86.94% accuracy. The model demonstrates accurate performance by delivering 99% precision alongside 95% recall, which minimizes defective detection errors. The model shows its excellence at detecting complex software defect patterns through experimental validation. The suggested method improves software quality by providing a data-driven, scalable solution.