A Classification Model for Software Bug Prediction Based on Ensemble Deep Learning Approach Boosted with SMOTE Technique

Abstract

In the software development process, the testing phase plays a vital role in assessing software quality. Limited resources pose a challenge in achieving this purpose efficiently. Therefore, early stage procedures such as Software Fault Prediction (SFP) are utilized to facilitate the testing process in an optimal way. SFP aims to predict fault-prone components early based on some software metrics (features). Machine Learning (ML) techniques have proven superior performance in tackling this problem. However, there is no best classifier to handle all possible classification problems.Thus, building a reliable SFP model is still a challenge and open for research. The primary purpose of this paper is to introduce an efficient classification framework to improve the performance of the SFP. For this purpose, an ensemble of Multi-layer Perceptron (MLP) deep learning algorithm boosted with Synthetic Minority Oversampling Technique (SMOTE) is proposed. The proposed model is benchmarked and assessed using sixteen real-world software projects selected from the PROMISE software engineering repository. The comparative study revealed that ensemble MLP achieved promising prediction quality on the majority of datasets compared to other traditional classifiers as well as those in preceding works.