Using Moodle Test Scores to Predict Success in an Online Course

Abstract

Many higher education institutions use a free, open-source learning management system (LMS) called Moodle that provides all the necessary tools for teachers to create virtual classrooms using the Internet. In this environment, teachers request reports detailing which course resources and activities learners accessed, including access times. Therefore, teachers can check individual student performance, whether students viewed a particular resource or participated in some activities in a certain period, based on reports known as log files. This paper deals with the analysis of data based on the scores of student online tests obtained using Moodle log files. Based of the collected data, which contain student data of the first year of undergraduate study Introduction to Programming at the University of Split, Faculty of Science and Mathematics in Croatia, pre-processing and export analysis of the data is carried out. Furthermore, prediction methods that include machine learning algorithms are used to predict student's final grade. It was shown that the highest accuracy of 82.35%, AUC and Cohen kappa with values of 0.766 and 0.706 were achieved with the Linear SVC algorithm for predicting student's performance. However, the challenge we faced is the lack of data, where three ways to achieve the most accurate performance model are presented. Also, we examined importance of considering significant attributes that influence student performance prediction results.