Analisis Siswa Bermasalah Menggunakan Algoritma K Means

Abstract

Problematic student identification in schools often relies on manual observation, fragmented counseling records, and subjective judgment, making early intervention less consistent. This study aims to analyze student problem patterns using the K-Means clustering algorithm so that schools can classify students into interpretable risk groups and design more targeted guidance services. The research used a quantitative data mining approach with an illustrative anonymized dataset consisting of 180 student records. Six attributes were modeled, namely unexplained absence, tardiness, disciplinary violations, average academic score, counseling record frequency, and task completion rate. The analytical process followed the CRISP-DM stages, including problem understanding, data preparation, normalization, determination of cluster number using elbow and silhouette evaluation, K-Means modeling, and interpretation of cluster profiles. The optimal structure was obtained at three clusters with a silhouette score of 0.58. The clusters were interpreted as low-risk monitoring students, medium-risk students requiring periodic assistance, and high-risk students requiring priority intervention. The high-risk cluster represented 18.9% of records and was characterized by higher absence, higher tardiness, more disciplinary violations, lower average scores, more counseling notes, and lower task completion. The findings indicate that K-Means can support school counseling by transforming administrative data into practical decision categories. The model should be validated with actual school data before operational use, but it provides a systematic framework for evidence-based student support.