Prediction of student learning style using modified decision tree algorithm in e-learning system

Abstract

Learning style is an important factor that accounts for the individual student learning in any learning environment. Every student has a different learning style and different ways to percept, process, retain and understand new information. In this paper, a new approach is proposed to classify students learning style automatically and dynamically depending on their learning behavior in a learning management system (LMS). There have been several approaches proposed for automatic learning style detection. One of the widely accepted and frequently used classification techniques is the decision tree classifier. The decision tree classifier mainly depends on the construction of strong decision rules which are required to identify learning styles accurately. The lack of strong decision rules would lead to the misclassification of individual students learning style. Hence, this paper mainly focuses on the construction of strong decision rules to strengthen the existing decision tree classifier to accurately and precisely classify students learning style thereby improving the classification accuracy. The proposed approach has experimented with an average of 300 students enrolled for the online courses in Moodle LMS. Initially, the students' behavior are extracted from the web log files of LMS and then preprocessed to build decision tree classifier using strong decision rules based on the three learning dimensions of standard Felder Silverman learning style model (FSLSM). The evaluation result is obtained using the inference engine with forward reasoning searches of the rules until the correct learning style is determined. Based on the result obtained, the prediction of learning style is done for the new students automatically and accurately using the significant rules built in the decision tree classifier. The experimental result proved that the processing dimension shows variance in classification whereas perception and input dimension shows less variance with an average accuracy of 87%.