An Ensemble Classification Approach Using Improvised Attribute Selection

Abstract

The advancement in data mining allows educational data to find patterns, improving the quality of the educational processes. Assessment of students' performance is of great importance for themselves as well as benefitting the educational institutes. In this regard, main attributes are generally proclaimed in the educational data mining (EDM) settings as a significant concern in learning analytics. In this research, we improved the prediction of students' performance. We evaluated the features containing six attributes in several domains: demographic, personal, academic, parental support, psychometric, and learning logs. The prediction of different classification algorithms using ensemble methods affects the accuracy and comprehensibility of the early prediction. To do so, we improvised attributes selection techniques applied to the data containing 11814 students in the biology course. The validation of the classification algorithms achieves better accuracy with ensemble methods. The result performances show the appropriateness of performing prediction and evaluating both filter and wrapper-based methods for feature selection. Our findings also show the students' performance with the most impactful features using ensemble methods and the feasibility of creating a prediction model with a reasonable accuracy rate.