偏相关方法在软件缺陷预测中的应用

摘要

In order to improve the performance of predictors, and reduce the dissention, this paper propsed a new predict model based on partial correlation analysis. Firstly, different to prior works, analyzed the correlation between attributes and defects. Then computed code complexity density values. Based on these values, built a new predictor. Experiments were performed on the public Eclipse dataset. This predictor had a good performance with high recall rates and substantially high F-measure values. The satisfactory results also confirm the partial correlation is a very important factor in software quality analysis. This conclusion is helpful for building more stable defect predictors.%为了提高预测模型的性能,解决不同属性子集带来的分歧,提出了基本偏相关方法的预测模型.首先,该方法在公开数据集上分析出代码静态属性与缺陷数之间存在偏相关关系；然后基于偏相关系数值,计算出代码复杂性度密度属性值；最后基于该属性值建立新的缺陷预测模型.实验表明,该模型具有较高的召回率和很好的F-measure性能,从而进一步证实了代码属性与模块缺陷之间的偏相关性是影响软件质量预测性能的重要因素的结论.该结论有助于建立更加稳定可靠的软件缺陷预测模型.