Research on software defect prediction technology based on deep learning

Abstract

To solve the problem that the traditional feature selection methods, such as PCA and LDA, are unable to get the nonlinear relationship between characteristics. Deep belief networks cannot eliminate the noise and missing value, which affect the accuracy of the software defect prediction (SDP) model. Not only the methods of feature selection, but data preprocessing and learning algorithm can also affect the precision of the defect prediction model. This thesis uses deep belief networks and SVM to construct an SDP model (DBN-SVM) to increase prediction precision. Using denoising autoencoders and SVM to build an SDP model (DA - SVM), compared with the DBN - SVM, DA - SVM model not only improves the prediction precision, but also enhance the robustness of the model. The thesis also proposes an SDP model framework which includes data preprocessing, feature selection and learning algorithm.