An Ensemble Model For Software Defect Prediction Using Machine Learning Algorithms

Abstract

The rapid growth of software development has brought along an increase in software defects, which can affect the quality, functionality, and performance of software applications. Software defects can result from various factors such as incomplete requirements, human errors, and the complexity of modern software systems. Since software development is time-consuming and costly, detecting and fixing these defects early is crucial. A primary method to control and reduce these defects is through software testing, particularly by testing each module in the software. However, manual testing alone cannot be fully effective in identifying defects, which is why defect prediction models have gained importance. The paper presents a comparative study of existing methods for software defect prediction by using various classification algorithms. The objective is to analyze how these algorithms perform in terms of accuracy, precision, recall, and other metrics when applied to software defect prediction tasks. Historical datasets, such as the NASA MDP (Metrics Data Program) datasets, which contain real-world defect data from NASA software projects, are used for training and testing these algorithms. By applying these datasets, researchers can evaluate the effectiveness of each algorithm in predicting defects. The proposed method, when compared with traditional classification algorithms, shows promising results, providing a new approach to improving software defect prediction.