Improving the cluster validity on student's psychomotor domain using feature selection

In the student clustering, the high cluster validity is very important because of this cause clarity a student in a cluster. Furthermore, it becomes easier for a teacher to do the best learning process. This paper focuses on the improvement of cluster validity applied by a suitable feature selection method, especially student's psychomotor domain. Here, we propose the feature selection by the random method. In addition, we apply k-means as the popular clustering method in educational data mining by the two initial of cluster center point: k-means++ and random. For cluster evaluation stage, silhouette coefficient is used on Manhattan distance. The experimental result indicates that feature selection is able to enhance the cluster validity which has shown that our methods have higher silhouette value than original k-means. In terms of the maximum silhouette value, our method can reach higher than original_kmeans++ and original_random on average 0.033–0.106. In terms of the minimum silhouette value, our method can achieve higher than original_kmeans++ and original_random on average 0.123–0.240.