A systematic review of software fault prediction techniques: models, classifiers, and data processing approaches

Abstract

Software fault prediction (SFP) plays a critical role in improving software reliability by enabling early detection and correction of defects. This paper presents a comprehensive review of 25 recent and significant studies on SFP techniques, focusing on data preprocessing strategies, classification algorithms, and their effectiveness across various datasets. The review categorizes the approaches into traditional statistical models, machine learning methods, deep learning architectures, and hybrid techniques. Notably, wrapper-based feature selection, neural network classifiers, and support vector machines (SVM) are identified as the most effective in achieving high accuracy, particularly when dealing with imbalanced or noisy datasets. The paper also highlights advanced approaches such as variational autoencoders (VAE), Bayesian classifiers, and fuzzy clustering for fault prediction. Comparative analysis is provided to assess performance metrics such as accuracy, F-measure, and area under the curve (AUC). The findings suggest that no single method fits all scenarios, but a combination of appropriate preprocessing and robust classification yields optimal results. This review provides valuable insights for researchers and practitioners aiming to enhance software quality through predictive analytics. Future work should explore ensemble learning and real-time SFP systems for broader applicability.