Detecting outsourced student programming assignments

Abstract

The task of writing computer programs outside of class is the most realistic experience students have in a programming class and hence can be the most accurate evaluation of their ability. However some students hire outside parties to produce these programs. We present a data mining and machine learning approach that can provide objective evidence for detecting such instances. Based on programs submitted by students across two lower-level CS (Computer Science) courses, we extract some basic programming style metrics. A decision tree model built on the collected measurements yields relatively good detection accuracy. In addition, an investigation into relative importance of the basic style metrics was performed which indicated Lines of Code, Number of Variables, and Number of Comments as important attributes. The methods are being implemented in a software analysis tool that instructors could possibly use for detecting outsourced program submissions.