Machine Learning and Student Activity to Predict Academic Grades in Online Settings in Latam

Abstract

In the past few years, interest in applying intelligent data-mining techniques to educational datasets has increased rapidly, with goals ranging from identifying students who need further support to being able to infer or predict a student’s final grade based on their behaviour during the learning process. Even more amongst students enrolled from all Latin America. This problem can be solved with solid technical approaches, but blind brute-force data analysis approaches may prove insufficient to accurately predict grades, and even if they managed, instructors may need to further understand why and how these algorithms predict specific grades. In this work, we use an experiment to better understand how different parts of the dataset influence the performance of different grade prediction algorithms. The goal is not to achieve the best possible prediction of student’s individual performance in an online university setting, with premises in half a dozen Latin American countries, and with Latin American students, but rather to identify which types of student activities are better predictors of the student’s actual performance. © 2021, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.