Enhanced Software Bug Prediction Using Double-Stacked Ensembles and Halving Search Optimization

Early detection of software bugs during the development process is vital for enhancing the quality of the software as well as controlling costs. In this research, we explore a novel method of Software Bug Prediction(SBP) using a Double-Stacking ensemble model. The tuning of this model through extensive hyperparameter optimization formed a major element of our work. After evaluating Grid Search, Random Search, and Halving Search we found Halving Search achieved the optimal balance between computational cost and prediction precision. To improve model robustness against the typical dataset imbalance present in SBP datasets, such as the NASA JM1 dataset with 10,885 instances and 21 attributes, we utilized data augmentation methods. Our empirical results emphasize the advantage of using the ensemble approach over single classifiers. Interestingly, the highly optimized Double-Stacking model achieved the highest accuracy of 86.6%, affirming that the integration of ensemble learning with sophisticated optimization and data augmentation yields a strong and efficient approach to high-performance SBP.