Multi-Layer Perceptron Neural Network with Feature Selection for Software Defect Prediction

Abstract

Software is continuously evolving and hence it is essential for the production of quality and stable software by every software provider. Recently there is a paradigm shift in how software is designed. One of the biggest challenges of software engineering is predicting defects in software modules, to save quality testing time. As software development challenges and constraints rise, unexpected effects such as failure and errors decrease the consistency of software and user loyalty, rendering error-free software more complex and frustrating. In this paper, we analyze the use of Multi-Layer Perceptron Neural Network [5] (MLP-NN) for the efficient prediction of defects. We have also executed the MLP-NN with a subset of features selected using popular feature selection methods. The model was evaluated on 5 datasets from the AEEEM dataset. The results were compared with other common classifiers like Logistic Regression, MLP-NN, and Random Tree. The findings indicate that feature selection has a major role in increasing the accuracy of prediction. Our model had higher accuracy in few cases while at par with others in some.