A Bayesian Ensemble Classifier for Source Code Authorship Attribution

Authorship attribution of source code is the task of deciding who wrote software, given its source code, when the author of the software is not explicitly known. There are numerous scenarios in which it is necessary to identify the author of a piece of software whose author is unknown, including software forensics investigations, plagiarism detection, and questions of software ownership. A number of methods for authorship attribution of source code have been presented in the past, including two state-of-the-art methods: SCAP and Burrows. Each of these two state-of-the-art methods was individually improved, and – as presented in this paper – an ensemble method was developed from them based on the Bayes optimal classifier. An empirical study was performed using a data set consisting of 7,231 open-source and textbook programs written in C++ and Java by thirty unique authors. The ensemble method successfully attributed 98.2% of all documents in the data set, compared to 88.9% by the Burrows baseline method and 91.0% by the SCAP baseline method.