COAT: Code Obfuscation Tool to Evaluate the Performance of Code Plagiarism Detection Tools

Abstract

There exist many plagiarism detection tools to uncover plagiarized codes by analyzing the similarity of source codes. To measure how reliable those plagiarism detection tools are, we developed a tool named Code ObfuscAtion Tool (COAT) that takes a program source code as input and produces another source code that is exactly equivalent to the input source code in their functional behaviors but with a different structure. In COAT, we particularly considered the eight representative obfuscation techniques (e.g., modifying control flow or inserting dummy codes) to test the performance of source code plagiarism detection tools. To show the practicality of COAT, we gathered 69 source codes and then tested those source codes with the four popularly used source code plagiarism detection tools (Moss, JPlag, SIM and Sherlock). In these experiments, we found that the similarity scores between the original source codes and their obfuscated plagiarized codes are very low; the mean similarity scores only ranged from 4.00 to 16.20 where the maximum possible score is 100. These results demonstrate that all the tested tools have clear limitations in detecting the plagiarized codes generated with combined code obfuscation techniques.