Explainable Machine Learning Models for Software Defect Prediction: An Empirical Study across Multiple Projects

Software defect prediction plays a critical role in proactive quality assurance by identifying fault-prone components early in the development lifecycle. While machine learning models have significantly improved predictive accuracy, their black-box nature limits trust and practical adoption in engineering environments. This paper proposes an explainable machine learning framework that integrates ensemble learning with SHAP-based interpretability to provide transparent and robust defect prediction across multiple software projects. The framework evaluates both baseline and advanced predictive models using cross-project validation to assess generalization capability. Experimental results demonstrate that ensemble models achieve superior accuracy while explainable analysis reveals consistent defect drivers related to software complexity, change frequency and historical fault patterns. Statistical significance testing confirms substantial performance improvements over traditional classifiers. The proposed approach bridges the gap between predictive performance and engineering interpretability, enabling actionable quality assurance decision support.