STUDENT DROPOUT RISK ANALYSIS USING A HYBRID K-MEANS AND CART METHOD

Abstract

Student dropout is an important issue in higher education because it affects study continuity, resource efficiency, and institutional performance. This study aims to develop a student dropout risk analysis model using a hybrid K-Means and Classification and Regression Tree (CART) approach within a two-stage modeling framework. The data were obtained from the Academic Information System of STMIK X for the academic years 2015/2016–2018/2019. After data cleaning and selection, 631 student records were obtained, consisting of 119 dropout students and 512 active/graduated students. In the first stage, K-Means was employed to identify the underlying student segmentation structure, while in the second stage, CART was applied using the cluster labels as an additional feature. The K-Means results identified K = 10 as the statistically optimal number of clusters, with a silhouette score of 0.6330. These ten clusters were subsequently interpreted into three risk groups: low, medium, and high. On the test data, the hybrid model achieved an accuracy of 92.25%, precision of 60.87%, recall of 87.50%, and an F1-score of 71.79%, outperforming the standalone CART model, which achieved an accuracy of 88.03%, precision of 44.44%, recall of 25.00%, and an F1-score of 32.00%. Validation using Orange Data Mining also demonstrated a consistent improvement, with the hybrid model achieving an accuracy of 92.60%. Attendance was identified as the most dominant predictor, with a feature importance value of 0.71. These results indicate that K-Means segmentation can enrich the data representation prior to CART classification and improve the early detection of students at risk of dropping out