An approach to educational data mining model accuracy improvement using histogram discretization and combining classifiers into an ensemble

Abstract

The paper presents an educational data mining model for predicting students’ final grades based on realized activities in different educational environments. The proposed model was generated through the stages of the data mining process: data set generation, preprocessing and application of appropriate classifiers. A training set was created by integrating multiple data sources. The concept of selecting appropriate discretization method and the classifier is based on a small training set consisting of different value domain data and a multidimensional class label. Accuracy of the proposed model was improved using unsupervised histogram discretization method and combining a classifier into an ensemble with majority voting algorithm. The unsupervised histogram discretization method reduced the effect of ignoring the class label. Significant results were achieved in individual class prediction using different classifiers. The contribution of the research presented in this paper is development of an efficient multidimensional class label prediction model for a blended learning environment case study.