Hybrid Machine Learning Approaches for Software Fault Prediction: A Systematic Review of Current Practices and Future Prospects

Software fault prediction (SFP) is a critical aspect of software engineering aimed at improving software reliability and reducing maintenance costs. Hybrid machine learning approaches, which combine the strengths of multiple models or techniques, have emerged as a promising solution to address the limitations of standalone methods in SFP. This systematic review synthesizes current practices in hybrid machine learning for SFP, analysing key methodologies, datasets, evaluation metrics, and their performance outcomes. The findings reveal that hybrid approaches leveraging ensemble learning, feature selection, and optimization techniques demonstrate superior fault detection accuracy and generalizability compared to traditional methods. However, challenges persist in areas such as model interpretability, dataset imbalance, and standardization of evaluation protocols. Based on the current landscape, this review identifies future research directions, including the integration of deep learning, transfer learning, and explainable AI in hybrid frameworks, to enhance predictive capabilities and practical adoption in realworld scenarios. This study provides a comprehensive foundation for advancing the field of hybrid machine learning in SFP.