AN IMPROVED CNN BASED ARCHITECTURE FOR WITHIN-PROJECT SOFTWARE DEFECT PREDICTION

Abstract

To improve the software quality, the software is generally tested to find out any bugs or a simple reliability test. A reliable software defect checking mechanism is a leading research topic, in the era of dependency on software’s for several tasks. Many researchers used different techniques of deep learning algorithm such as CNN i.e convolutional neural networks, and deep belief networks for prediction of software defect. these algorithms failed to provide higher prediction accuracies. To overcome the issues a new algorithm for software defect prediction is required for higher accuracy and other parameters like F- and G-measure and specifically important parameter is Matthews correlation coefficient (MCC) measure. In this paper, a new modified CNN algorithm is proposed, which combines the CNN based models into one and apply concatenate algorithm under SVM i.e. support vector machine classifier. The results clearly indicate that the proposed algorithm improves the parameters and thus is a highly dependable and reliable method for software defect prediction.