MACHINE LEARNING TECHNIQUES FOR SOFTWARE DEFECT PREDICTION: A SYSTEMATIC LITERATURE REVIEW

Abstract

Developers and organizations are facing challenges to ensure software quality because software complexity is growing rapidly. Software defect prediction is used to predict defects early by using machine learning techniques on historical data in the development process.  Early defect prediction improves the accuracy, quality, and reliability of software. It also reduces overall cost and time consumption. This paper includes a systematic literature review of machine learning methods used in software defect prediction. A total of 15 research articles published in the past 10 years were examined and selected from reliable resources, including IEEE Xplore, Science Direct, and Google Scholar. The research evaluates and compares the performance, pros, and cons of many machine learning techniques, such as Naive Bayes, Decision Trees, Artificial Neural Networks, Random Forest, and Support Vector Machine. Many studies claim that Random Forest has achieved the most significant prediction of accuracy among all studied techniques because it can efficiently handle noisy and imbalanced data. On high-dimensional datasets, Support Vector Machine works efficiently while Naive Bayes is effective for limited data resources. However, the model's reliability remains influenced by factors such as high computational power, data quality, and limited cross-project generalization. Thus, machine learning plays an important role in improving the quality, reliability, and accuracy of software defect prediction.