Software forensics for discriminating between program authors using case-based reasoning, feedforward neural networks and multiple discriminant analysis

Abstract

Software forensics is the field that, by treating pieces of program source code as linguistically and stylistically analyzable entities, attempts to investigate computer program authorship. This can be performed with the goal of identification, discrimination, or characterization of authors. In this paper we extract a set of 26 standard authorship metrics from 351 programs by 7 different authors. The use of feedforward neural networks, multiple discriminant analysis, and case-based reasoning is then investigated in terms of classification accuracy for the authors on both training and testing samples. The first two techniques produce remarkably similar results, with the best results coming from the case-based reasoning models. All techniques have high prediction accuracy rates, supporting the feasibility of the task of discriminating program authors based on source-code measurements.