Research on Preprocessing Techniques for Software Defect Prediction Dataset Based on Hybrid Category Balance and Synthetic Sampling Algorithm

With the popularization of computer software in various fields, software defects have become a hot topic of industry concern. Traditional testing methods typically require a significant amount of resources and time, however, due to insufficient testing coverage, potential software defects are often not detected in a timely manner. In practice, imbalanced dataset categories, excessively high feature dimensions, and limitations of single classifier prediction accuracy still pose challenges in software defect prediction. To address these challenges, this paper proposes a preprocessing strategy based on mixed category balancing and synthetic sampling. This strategy removes noisy samples through undersampling and uses oversampling methods to solve the problem of imbalanced data categories, thereby improving the quality of data. In addition, to overcome the problem of limited accuracy of a single classifier, a multi model weighted voting fusion strategy is proposed. By integrating three base classifiers, support vector machine, K-nearest neighbor, and random forest, different weights are assigned based on the accuracy of each base classifier during the decision-making stage, and a weighted voting mechanism is used to generate the final prediction results, significantly improving the accuracy of software defect prediction. By combining data preprocessing methods and weighted voting algorithms, this paper constructs a more accurate software defect prediction model and verifies its superiority in open source projects. We also designed and implemented a software defect classification and prediction system based on this model, which helps testers optimize the prediction process and provide support for decision-making.