Defect Prediction in Object-Oriented Software with Decision Tree Classifiers

Abstract

Object-oriented defect prediction helps software teams identify classes that require stronger testing, inspection, or refactoring before defects spread into later development stages. This article examines Decision Tree and Naive Bayes classifiers for predicting defective object-oriented software modules using class-level design metrics related to size, coupling, cohesion, inheritance, response behavior, interface exposure, and change history. The framework prepares defect labels, screens dataset quality, handles imbalance cautiously, and compares both classifiers through error-cost-oriented measures rather than accuracy alone. The study shows that Naive Bayes is more suitable when the main priority is reducing missed defective classes, while Decision Tree is more useful when review effort must be controlled and interpretable defect rules are required. The proposed approach supports practical defect prediction by linking classifier choice with testing capacity, defect escape risk, and maintainability planning in object-oriented software projects.