Dhaka University Journal of Applied Science & Engineering

Software defect prediction is a key research area in the domain of software quality estimation. Usually, software attributes are used for building a defect prediction model and a specific prediction model can produce positive, negative, or neutral outcomes depending on the characteristics of these attributes. Therefore, choosing an optimal set of attributes for the development of a defect prediction model remains a vital yet relatively unexplored issue. To address this issue, we propose a technique for attribute selection to improve the accuracy of software defect prediction for both within project and cross-project. Experimental results using the data sets from Relink and NASA MDP repository demonstrate the superiority of the proposed algorithm

1. ref1