A Framework for Software Defect Prediction Using Optimal Hyper-Parameters of Deep Neural Network

Abstract

Software defect prediction (SDP) models are widely used to identify the defect-prone modules in the software system. SDP model can help to reduce the testing cost, resource allocation, and improve the quality of software. We propose a specific framework of optimized deep neural network (ODNN) to develop a SDP system. The best hyper-parameters of ODNN are selected using the stage-wise grid search-based optimization technique. ODNN involves feature scaling, oversampling, and configuring the base DNN model. The performance of the ODNN model on 16 datasets is compared with the standard machine learning algorithms viz. Naïve Bayes, Support Vector Machine, Random Forest, Ada Boost, and base DNN model in terms of Accuracy, F-measure, and Area Under Curve (AUC). Experimental results show that the ODNN framework outperforms base DNN (BDNN) with 11.90% (accuracy), 0.26 (f-measure), and 0.13 (AUC). The statistical analysis using Wilcoxon signed-rank test and Nemenyi test show that the proposed framework is more effective than state-of-the-art models. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.