Random wheel: An algorithm for early classification of student performance with confidence

Abstract

The educational data mining researchers have achieved significant efficiency in predicting student performance during the tenure of the course. However, an early prediction before course commencement is still a research challenge. Such advanced forecast can help the teachers in providing timely assistance to uplift the academic performance of a student, reduce the number of failures and performance degradations. Importantly, an additional measure of prediction confidence can be useful in this regard to decide the magnitude of the assistance required. The primary objective of this study is to predict the failure, degradation and improvement before course commencement. A real dataset containing nearly 0.6 million records is used here for this purpose. We have initially applied multiple state-of-the-art classifiers on this dataset to predict the performance in binary terms. Unfortunately, these classifiers could not perform well, and they are unable to provide the desired prediction confidence as well. We have therefore proposed a novel scalable algorithm, named random wheel, for classification. It not only works efficiently on this dataset but also works well with other benchmarked datasets. The proposed classifier provides an additional measure to indicate the prediction confidence. It, in turn, increases the acceptability of the prediction.