一种基于Logicboost的软件缺陷预测方法

Software Defect Prediction Method Based on Logicboost

针对软件缺陷预测中对不平衡数据分类精度较低的问题,提出了一种新的基于LogitBoost集成分类预测算法,使用SMOTE方法对原始数据集进行平衡处理,然后使用随机森林算法作为弱分类器算法进行分类预测,最后使用LogitBoost算法以加权方式集成各弱分类器的结果。通过在NASAMDP基础数据集上验证得出本文提出的分类预测算法比数据集均衡处理前准确率高出0.1%-11%,同时在均衡处理后比KNN算法平均高出0.9%,比SVM算法平均高出0.4%,比随机森林算法平均高出0.1%。

Aiming at the problem of low classification accuracy of unbalanced data in software defect prediction, a new integrated classification prediction algorithm based on LogitBoost is proposed. SMOTE method is used to balance the original data set, then random forest algorithm is used as weak classifier algorithm for classification prediction. Finally, the results of weak classifiers are integrated in a weighted way using LogitBoost algorithm. Through the verification on NASA MDP basic data sets, the classification prediction algorithm proposed in this paper is 0.1%-11% higher than that before data balancing, 0.9% higher than that of KNN algorithm, 0.4% higher than that of SVM algorithm and 0.1% higher than that of random forest algorithm.