Predicting Students’ Behavior During an E-Learning Course Using Data Mining

This paper introduces a process of building a prediction model for student’s final grade and time of finishing, based on students’ previous behavior. Prediction model was developed using data mining with regression analysis, principle component analysis and hierarchical clustering of symbolic histogram valued data. 35 different features of students’ activates was considered but only the 9 most important, so called principle components, were used in the model. Then, using histogram valued data - a type of symbolic data that allows learning processes to be described in a more natural form, and a hierarchical clustering, previous students’ behaviors were grouped. For an accurate prediction, a closest cluster to student’s current progress was found. To verify the model’s correctness, predictions were tested on a largest course in e-learning system in 2015 fall semester. The model was found to work sufficiently.