A Comparative Analysis for Machine Learning based Software Defect Prediction Systems

Abstract

In the Software Engineering concept, the prediction of the software defects plays a vital role in increasing the quality of the software systems, which is one of the most critical and expensive phases of the software development lifecycle. While the use of software systems is increasing in our daily lives, their dependencies and complexities are also increasing, and this results in a suitable environment for defects. Due to the existence of software defects, the software produces incorrect results and behaviors. What is more critical than defects, is finding them before they occur. Therefore detection (and also prediction) of the software defects enables the managers of the software to make an efficient allocation of the resources for the maintenance and testing phases. In the literature, there are different proposals for the prediction of software defects. In this paper, we made a comparative analysis about the machine learning-based software defect prediction systems by comparing 10 learning algorithms like Decision Tree, Naive Bayes, K-Nearest Neighbor, Support Vector Machine, Random Forest, Extra Trees, Adaboost, Gradient Boosting, Bagging, and Multi-Layer Perceptron, on the public datasets CM1, KC1, KC2, JM1, and PC1 from the PROMISE warehouse. The experimental results showed that proposed models result in proper accuracy levels for software defect prediction to increase the quality of the software.