A Comprehensive Analysis of Unified Approaches for Revealing Code Clone Detection

Code clone detection can be crucial in software maintenance, because of clone occurrences have potential of causing errors, raise costs, or results in unsynchronized modify. Analyzing the limitation of the existing clone detection method, we found that while the token and AST thin methods have been rather successful at detecting simpler clone types, they are significantly inadequate when used for identifying Type III and Type IV clones. The latter are especially significant in cross-language cases in which mere conjunctions of syntactic predicates are insufficient. This paper introduces an enhanced method for code clone detection through combining token-based, abstract syntax tree based, program dependency graph based, and machine learning based clone detection methods in a single framework. In this paper, we developed a scalable approach using adaptive ensemble weighting and fine-tuned LLMs that improved the precision over all types of clones together with low FPs in both languages. The selective execution model of the proposed approach also improves resource utilization making it possible to work on large codebase. Experimental outcomes prove that the proposed approach has higher accuracy, precision, and less computation time compared with the state-of-the-art techniques for clone detection, which make it a valuable contribution for clone detection study.