Software Defect Prediction using Deep Learning

Abstract

An increasing number of defects in software, damages the quality and reliability of that software.The detection of defective instances is becoming increasingly important, and current detection techniques require a great deal of improvement.However, Machine Learning (ML) techniques are effectively used, to detect defects in software.The primary purpose of ML techniques in Software Defect Prediction (SDP) is to predict defects, according to historical data.Establishing a critical SDP model on high-dimensional and limited data is still a challenging task.Thus, in this paper, we proposed an approach to detect defective modules in software using enhanced Convolutional Neural Networks (CNNs).The paper aims to identify the defective instance using the enhanced deep learning method.Our experiments are based on Within Project Defect Prediction (WPDP), where K-fold cross-validation is performed.The proposed approach has been evaluated on nineteen opensource software defect datasets, with respect to different evaluation metrics.Empirical results show that our proposed approach is significantly better than Li's CNN and standard ML model.In addition, we performed the Scott-Knot ESD test, which shows the effectiveness of our proposed approach.