A Study on the Significance of Software Metrics in Defect Prediction

Abstract

In the case of metrics-based software defect prediction, an intelligent selection of metrics plays an important role in improving the model performance. In this paper, we use different ways for feature selection and dimensionality reduction to determine the most important software metrics. Three different classifiers are utilized, namely Naïve Bayes, support vector machine and decision tree. On the publicly NASA data, a comparative experiment results show that instead of 22 or more metrics, less than 10 metrics can get better performance.