A Novel Clustering Based Feature Selection for Classifying Student Performance

The main intent of this work is to identify and eradicate the irrelevant as well as redundant features that are used to improve the accuracy of student performance classification.

In this article, a novel technique is introduced for feature or attributes selection purpose called as Non-negative Matrix Factorization Clustering based Feature Selection (NMFCFS). NMFCFS uses symmetric uncertainty (SU) estimation.

The performance of this work is evaluated by using the student dataset that includes collection of students' information from various colleges. For analyzing the performance of this work, comparative evaluation is performed between the classifiers (in the experiment, Prism and J48 is taken) without feature selection and classifiers with the NMFCFS. The experimental result shows that NMFCFS approach is attaining higher accuracy rate i.e., 97.8%. The proposed feature selection method is highly efficient when compared to other schemes.

The findings demonstrate that the proposed method has high performance of the students' failure and dropout prediction. In other words, this can improve the accuracy of the classification result.