Model for Approximating Students' Performance in Programming Exercise Lecture Using Multi-Agent Simulation

Abstract

Early prediction of students with learning problems is one of the challenges in the field of education. Students who have problems studying are those who have difficulty in earning credits because of their low test scores and their reluctance to participate in classes. In order to provide appropriate support measures to students with learning problems, it is important to correctly predict student performance based on learning-related data. In this research, we propose an approximate model for the transition of the achievement level of the comprehension test conducted in the lectures for the programming exercises offered in our university. Specifically, we use the Agent Based Model (ABM) to construct an approximate model. Note that the ABM used in this study is simplified without considering the interaction between agents, which is a feature of standard multi-agent simulations. Then, we compare the simulated results by ABM with the average percentage of correct answers and variance of the correct answer data to verify the validity of the proposed model and evaluate the approximation accuracy. As a result, the coefficient of determination of the proposed model is 0.893, which is a high approximation accuracy.