Student intervention system using machine learning.

Abstract

In most universities, advisors guide students in their course selection, and warn those who might be at risk of being dismissed or placed on probation. The large number of students makes it difficult for universities to identify those at risk, as it would get very time consuming and inaccurate. Hence, there is a need for a system that can recognize these students at the end of each semester. In this work, we build a semester-based automated intervention system using machine learning that identifies students who are at risk and suggests upcoming courses, providing them with a more personalized approach. We applied supervised learning such as logistic regression, neural networks, and AdaBoost to predict three outcomes: 1) risk of dismissal, 2) risk of probation, and 3) time needed to graduate. Then we applied reinforcement learning (RL) using value iteration technique to create an optimal policy that will recommend courses to students with the goal of keeping them safe. We applied different evaluation methods, such as ROC curve and F-measure, to compare the performance of the supervised learning algorithms, and proposed a new approach to measure the effectiveness of the recommendation system. The dataset was provided by the American University of Beirut and contained a sample of 30,000 student records.