Comparative analysis for boosting classifiers in the context of higher education

Abstract

Machine learning techniques are applied on higher education data for analyzing the interaction between the students and electronic learning systems. This type of analysis serves in predicting students' scores, in alerting students-at-risk, and in managing the degree of student engagement to educational system. The approaches in this work implements the divide and conquer algorithm on feature set of an educational data set to enhance the analysis and prediction accuracy. It divides the feature set into a number of logical subgroups based on the problem domain. Each subgroup is analyzed separately and the final result is the combination of the results of the analysis of these subgroups. The classifier that shows the best prediction accuracy is dependent on the logical non-statistical nature of the features in each group. Both traditional and boosting classifiers are utilized on each dataset, from which a comparison is conducted to show the best classifiers along with the best dataset. This approach provides the possibility to apply a brute force algorithm in the selection of the best feature subgroups with a low computational complexity. The experimental work shows a high prediction accuracy of the students-at-risk relative to the current research, and provides a list of new important features in the field of electronic learning systems.