Software defect prediction model based on improved twin support vector machines

Software defect prediction contributes to ensuring the quality of software development and reducing software maintenance costs. However, the class imbalance problem can affect the accuracy of defect prediction classification, which is a crucial issue to be solved urgently. We propose a novel software defect prediction model based on a twin support vector machine to address imbalanced data classification issues and optimize the prediction effect. The model embeds the within-class structure of the training samples as the regularization term into the objective function, considering the structural information hidden in the data, and obtains the class structure information through clustering. Moreover, by introducing within-class structure information to maximize the within-class distances and one class intervals, the model produces a superior classification hyperplane and enhances the generalization ability of the support vector machine. The experimental results demonstrate that the proposed algorithm achieves higher prediction accuracy, more robust adaptability, and optimized performance in classifying imbalanced data compared with existing algorithms.