Emperical study of defects dependency on software metrics using clustering approach

Defect Prediction prior to the release of software uses metrics and fault data to know which properties of software are associated with faults in classes. In this paper, predication of software defects have been performed with the help of static code metrics. The proposed approach analyzed by multiple regression technique. Initially all the collected metrics are grouped in similar category using the K-means clustering approach results in more similar metric are in one cluster. The clustering performed based on structural information provided by collected data sets. In next step, empirically the impact of defect count metric on different clusters has been identify using regression approach. Finally the regression results shows prediction rate for defect count by each cluster. The result conclude the prediction model developed on clustering totally outperform those models that use only static metrics.