The software defect can damage the reliability and the quality of the software. The static code software metrics have been widely used and played an important role in software defect prediction. Instead of using whole features, it is quite necessary to remove the redundant features and select some meaningful features to improve the prediction performance. This study focuses on the effective attribute selection technique for the software fault classification. We proposed the software attributes network that indicates the mutual information between features and the clustering based attribute selection techniques. The results demonstrate that the proposed network clustering based feature selection performs the best on fault-prone modules prediction. The comparative feature selection techniques are examined to evaluate the result. Furthermore, the best-performed software attributes and the relations between them are shown and carefully analyzed.
展开▼
