Different learning predictors and their effects for Moodle Machine Learning models

In this paper 16 different Moodle Machine Learning models for predicting the success of 57 full-time students enrolled in the Applied Statistics course at the University of Dunaújváros in Hungary have been developed and tested in terms of “goodness”. The success can be affected by several factors, but here only students' cognitive activities are examined. The predictors used in the models are based on: number of view of PDF lecture notes, number of views of video lectures, number of views of books of solved exercises, number of views of Minitab videos (videos for problem solving with a statistical software), number of attempts of quizzes and best grades achieved by students on quizzes. The models differed in the number and in the types of predictors. Binary Logistic Regression was used for model training and evaluation. The target of the models indicates whether a student is at risk of not achieving the minimum grade to pass the course. The impact of cognitive predictors that are part of the Moodle core Analytics API on predictive power was also examined. Having evaluated the goodness of the different models, it was shown that students' success can be predicted purely from cognitive activities, but their predictive powers are very diverse. The predictors of quizzes have the largest impact on the success, however, supplementing the model with other even less effective predictors much better model can be made. Models built from purely Moodle core cognitive predictors give much less reliable results.