A novel ensemble feature selection and software defect detection model on promise defect datasets

Abstract

The prediction of software defects is an essential step before building high quality software. Although much research has been done for analyzing the software metrics and feature extraction. Unfortunately, traditional models failed to predict the defects using the multiple software projects data. As the number of software projects and modules increases, the sparsity and uncertainty of the data increases, which affects the overall true positive rate of the defect prediction process. In this paper, a hybrid ensemble feature selection and defect prediction model was designed and implemented on the openscience software defect dataset. ReliefF, Chi-square and improved predictive correlation measures are used in our ensemble feature selection process. Experimental results show that proposed model has high defect detection rate, recall and F-measure compared to the traditional software defect prediction models.