<![CDATA[Software quality depends heavily on the early detection of potentially defective modules, yet the complexity of software metrics and class imbalance often leads to inconsistent prediction performance. This study aims to compare the effectiveness of Radial Basis Function Neural Network (RBFNN) and RBFNN with Dynamic Decay Adjustment (RBFNN-DDA) in predicting software defects using five NASA PROMISE datasets (CM1, KC1, MC1, MW1, and PC1). The research employed quantitative experimentation through data normalization, a 70 to 30 train–test split, and model evaluation across maximum iterations ranging from 200 to 1,000. Model performance was assessed using Accuracy, Precision, Recall, F1 Score, and AUC. The results indicate that RBFNN provides higher Recall and F1 Score, making it better at identifying defective modules, although its performance is less stable. Meanwhile, RBFNN-DDA yields more consistent performance with higher Precision, Accuracy, and AUC on imbalanced datasets, albeit with lower Recall. Both models reached performance saturation at 200 until 400 iterations, showing minimal improvement at higher iteration counts. The findings imply the need for balancing sensitivity and stability when selecting defect prediction models, particularly in environments with severe class imbalance]]>
Copyrights © 2025
