Interpretable SHAP-Driven Machine Learning for Accurate Fault Detection in Software Engineering

In order to design and develop secure and reliable software, accurate prediction of software errors is crucial. The complex, nonlinear relationship between software features and bugs that occur during development despite precautions and measures taken to prevent them makes it difficult for traditional empirical models to predict these bugs with any degree of accuracy. An integration between the Decision Tree (DT) model and the SHAP (Shapley Additive exPlanations) technique was developed in this work with the aim of predicting software faults and providing informative explanations of the expected results. The synergistic advantages of integrating DT and SHAP allow the creation of an accurate, efficient, and fully interpretable technique. In order to make the process visible and reliable, SHAP provides a global explanation of how features of the developed software affect quality and a local explanation of how features contribute to each prediction. The tendency of software features to influence software quality was revealed by feature dependence analysis.