EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR STUDENT PROFILING IN ONLINE JUDGE ENVIRONMENTS

Abstract

Because they provide quick and impartial evaluations of the code that students write, online judge (OJ) systems are frequently taken into consideration in programming-related courses. Based on a rubric, such an evaluation often yields a single conclusion, usually indicating whether the submission completed the task satisfactorily. Nevertheless, it would be advantageous for the student and the teacher to get more feedback regarding the task's overall progress, as such data can be considered inadequate in an educational setting. By taking into account the potential for future utilisation of the data collected by the OJ and automatically deriving feedback for both the teacher and the student, this study attempts to address this issue. More specifically, we examine the modelling of student behaviour using learning-based schemes, including Multi-Instance Learning and conventional Machine Learning formulations. Additionally, Explainable AI is being considered to give feedback that is intelligible to humans. The concept was assessed using a case study that included 2,500 entries from about 90 different students enrolled in a computer science degree course that dealt with programming. The outcomes gained support the proposal: based just on the behavioural pattern deduced from the submissions made to the OJ, the model can accurately anticipate the user outcome (passing or failing the assignment). Additionally, the proposal may pinpoint student groupings and profiles that are more likely to fail as well as other pertinent data, which ultimately provides feedback to the teacher and the student.