Feature analysis for predicting students' performance from reading patterns in an e-learning system

Abstract

Final grade scores of students were predicted based on how they accessed teaching content provided by an e-learning system, BookRoll. In order to train machine learning models, features were designed heuristically, and various machine learning methods were trained and compared. The result showed that random forest and AutoML performs well. Analyzing trained random forest predictors revealed that time-related features contribute significantly to the performance of the regressor.