A Just-In-Time Software Prediction Model Based On A Double-Weighted Adaboost

Just-in-time software defect prediction can not only ensure software quality in the development process but also enable developers to check and repair defects on time. Instant defect prediction has the advantages of fast speed and easy tracking, but it is susceptible to class imbalance. In addition, the dataset contains many redundant and irrelevant features, which increases the complexity of the model and reduces prediction accuracy. To improve the performance of software defect prediction, we attempt to overlay ensemble learning and propose a Double-Weighted AdaBoost Just-In-Time software defect prediction framework DWA_JIT, which utilizes various classification algorithms to improve the performance of just-in-time software defect prediction. In the feature selection stage, DWA_JIT adopts the method of randomly selecting feature vectors based on a percentage for each base classifier to reduce data dimensions and improve model accuracy; In the classification stage, multiple AdaBoost frameworks are iterated through the internal base classifiers, and the external multiple AdaBoost are iterated again to form the final dual weighted AdaBoost classification model. The results show that the framework proposed in this article has good classification performance on software defect datasets, and has significantly improved the performance compared to the baseline model.