A Student Performance Predication Approach Based on Multi-Agent System and Deep Learning

Abstract

Educational Data Mining (EDM) has been a popular research topic in education, and many current studies use EDM techniques to predict student performance, so that the teachers and students can understand the student's performance in real-time and further develop the learning plan for the students. However, current work is often not sufficiently accurate in predicting student performance. Firstly, the students' features are not adequately processed, resulting in a large amount of noise data in the student dataset, affecting the prediction results. Secondly, there is still space for improvement in the current studies on the student performance prediction model. Therefore, in this paper, the Multi-Agent System (MAS) idea is used to propose an Agent-based Modeling Feature Selection(ABMFS) model, and the selected feature subset effectively removes the features that are irrelevant to the prediction results. Next, the Deep Learning techniques are used to construct a Convolutional Neural Network (CNN) based structure to predict student performance. The result of the experiments shows that the ABMFS Model selects the targeted features and improved performance noticeably across different classifiers, and better prediction results are achieved when the proposed approach was used for student performance prediction.