Validating and Detecting User-Specific Code Clones: An AI Framework Leveraging Metric-Based Feature Vectors

Abstract

Like other verification aspects, code clone validation remains highly subjective and user-dependent. This research presents an AI-based approach utilizing fragment-specific metric-based feature vectors to identify and validate customized code clones. We derive classification feature vectors through appropriate code metrics, training various machine learning models for identifier classification. The resulting framework enables users to submit code clone pairs for cloud-based validation. Upon submission, the trained AI model analyzes pairs using their metric features, generating user-specific validation scores returned via a RESTful API. We describe the framework architecture encompassing metric extraction, model training, and cloud deployment. Experimental results demonstrate the framework's ability to adapt effectively to individual validation strategies, optimizing accuracy while significantly reducing inspection effort compared to non-customized clone detection systems. A prototype system demonstrates the feasibility of providing automatically computed AI-based validation scores integrated with existing validation tools.