Identifying "At-Risk" Students: An AI-based Prediction Approach

Abstract

Student retention is of the utmost importance to higher education institutions. It is a metric used by legislators, accreditation agencies, and governing bodies. Providing students with remedial assistance at the right time has often proven an effective method for student retention. Identifying students that require this type of support is usually cumbersome though. A variety of stakeholders, such as educators, counselors, advisors, and other staff members, may have to be involved in identifying students who are “at-risk”. Following recent developments in machine learning algorithms, automated systems may be developed to predict students' performance and refer students to remedial instruction. This paper proposes the utilization of Artificial Intelligence (AI) algorithms to predict a student's grades in a university course at any given semester based on the initial performance of the student in a combination of course assessment tools, such as quizzes, assignments, and tests. The prediction model is based on a dataset of real cases compiled from courses at a private university in Saudi Arabia. The model, however, is general enough to be applied to any course at universities around the world. The prediction classifiers used in this study are Random Forest (RF), Sequential Minimal Optimization (SMO), Linear Regression (LR), Additive Regression (AR), and Multilayer Perceptron (MLP). Various metrics are employed to measure the prediction models' performance and assess the accuracy and validity of the proposed AI-based algorithms. Results indicate that the best classifier for predicting the final exam grade is the SMO, with a minimum mean absolute error of 2.350. The best prediction classifier for the midterm exam is LR with a minimum mean absolute error of 1.978. As tools for the early identification of students’ difficulties in the particular courses in which they are enrolled, the effectiveness of the proposed models is discussed.