Abstract: Predicting defect-prone software modules accurately and effectively is an important way to control the quality of a software system during software development. Feature selection can highly improve the accuracy and efficiency of the software defect prediction model. A mutual information-based feature selection method for software defect prediction is proposed. The optimal feature subsets generated by the proposed approach were applied to train and test various prediction models. The experiment results show that all the classifiers achieve higher accuracy and performance by using the feature subset provided by proposed approach.

Key words: Software quality, Mutual information, Feature selection, Optimal feature subset, Defect prediction

摘要： 软件开发过程中准确有效地预测具有缺陷倾向的软件模块是提高软件质量的重要方法。属性选择能够显著地提高软件缺陷预测模型的精确度和效率。提出了一种基于互信息的属性选择方法，将选择出的最优属性子集用于软件缺陷预测模型。方法采用了前向搜索策略，并在评价函数中引入非线性平衡系数。实验结果表明，基于互信息的属性选择方法提供的属性子集能提高各类软件缺陷预测模型的预测精度和效率。

关键词: 软件质量, 互信息, 属性选择, 最优属性子集, 缺陷预测