融合加权代码异味强度因子软件缺陷预测模型

Software defect prediction model integrating weighted code smell intensity

现有代码异味强度检测方法未考虑度量对代码异味的影响力度。基于已有代码异味强度检测方法,分析度量对代码异味的影响力度,改进代码异味强度检测方法。应用随机森林得到影响代码异味的度量的特征重要性,将各度量的评估值作为各度量的权值,检测加权代码异味强度的值。评估检测策略时,对各个度量的相关性进行分析,发现用于检测的度量之间缺乏相关性,符合从不同角度衡量代码异味的思想。当加权代码异味强度作为缺陷模型预测因子时,可提高缺陷预测模型约2%的F-Measure值。

The existing code smell intensity detection methods fail to consider the impact of measurement on code smell.Based on the existing code smell intensity detection methods,the impact of measurement on code smells was analyzed,and the code smell intensity detection method was improved.The characteristic importance of metrics affecting code smell was calculated using random forest.The evaluation value of each measurement was used as the weight of each measurement to detect the value of weighted code smell intensity.It is found that there is a lack of correlation between the measures used for detections,which is in line with the idea of measuring code smell from different angles.When the weighted code smell intensity is used as the predictor of the defect model,the F-measure value of the defect prediction model can be increased by about 2%.