Substantiation of Software Defect Prediction using Statistical Learning: An Empirical Study

Defect Prediction plays a very important role in the development of the software. The defect prediction leads to a more focused testing thereby reducing the testing cost and efforts. In this paper an empirical study is done to investigate the predictive performance of the statistical model for defect prediction. Three open source software projects are taken for the experiment. The Select K Best algorithm is used for feature selection. Random Forest, Gradient Boosting, K Nearest Neighbors, Decision Tree and Naïve Bayes are used as the statistical models. The data models are analyzed using various performance measures. The findings of this study infer that Select-K-Best with chi-square score function with the K Nearest Neighbors gives the optimal performance for software defect prediction.