A Review of Software Defect Prediction Models using Machine Learning

Abstract

One of the biggest issues in software development is software defects which greatly influence reliability, performance, and cost of the system used. The issue of undetected defects can result in system breakdowns, security breaches, and dissatisfaction among the users, hence early defect detection is a paramount activity in software engineering. Conventional methods of defect detection, e.g., manual code inspection, rule-based testing, are time-consuming and expensive to use and are constrained by the complexity of large software systems. Due to this, automated and intelligent methods have received more and more interest. Machine learning (ML) has become a useful tool in predicting defects in software by studying the past trends in software data and identifying modules that are likely to have faults during operation. ML-based models have a wide range of software metrics such as code, object-oriented, and process metrics that help to detect possible defects more accurately and efficiently. Decision Trees, Random Forests, Support Vector Machines and Naive Bayes are some of the broadly used supervised learning techniques, and recent research also investigated deep learning models to learn more complex and nonlinear software data relationships. The purpose of this review is to analyze in detail the currently available software defect prediction models using machine learning methods. This study will cover the analysis of the most popular datasets, preprocessing techniques, feature selection approaches, and performance measurement metrics. The most important findings reveal that the ensemble and hybrid machine learning model tend to be more successful in defect prediction tasks compared to single classifiers. However, it is important to note that issues like the imbalance of the classes, the dependence on the dataset and the inability to understand the model are also significant research questions. The existing trends in research focus on deep learning, cross-project defect prediction, and explainable artificial intelligence to advance the reliability of predictions and their real-world usefulness.

Keywords: Software Defects, Software Defect Prediction, Machine Learning Techniques, Software Metrics, Ensemble Learning, Deep Learning, Predictive Software Quality.