Educational Data Mining Using Semi-Supervised Ordinal Classification

Abstract

Until now, many different data mining techniques have been used for classification in the field of education. The main problems associated with this issue are insufficient labeled data and classification using nominal class labels, rather than ordinal ones. To overcome these problems, in this study, the Semi-Supervised Ordinal Classification (SSOC) method was tested on education data and satisfactory results were obtained. The SSOC method combines both semi-supervised learning and ordinal classification approach. Thanks to this method, a significant increase in accuracy achieved on three different education data. In the experiments, the SSOC method was tested by using different base learners, including Decision Tree, Random Forest, and Neural Network. For semi-supervised learning, the classification results were obtained with different quantities of labeled instances varying from 15% to 50% with 5% increments and finally as 75% labeled data. The datasets used in this study have an ordinal categorical structure such as $(c1 \lt c2 \lt c3)$. The experimental results show that the SSOC method can achieve satisfactory performance in case of using data that has ordinal structure but has small amount of labeled instances and large amount of unlabeled instances. In the field of education, it is highly likely that datasets related to student achievement have ordinal structure. Therefore, SSOC can be successfully used in classification studies in education.