<![CDATA[Purpose – This study examined relative academic-score profiles among sixth-grade students at SD Negeri 05 Bilah Barat using Educational Data Mining and K-Means clustering. It compared candidate cluster solutions using the Elbow Method, Silhouette evaluation, repeated-run stability, and educational interpretability. Design/methodology/approach – First-semester report-card scores from 50 students across Civic Education, Indonesian Language, Mathematics, Natural Sciences, and Social Sciences were Min–Max normalized and processed through one Python-based pipeline. K-Means used k-means++ initialization, 100 initializations, a fixed random seed of 42, a maximum of 300 iterations, Euclidean distance, and the Lloyd algorithm. Candidate solutions were evaluated using WCSS, overall and cluster-specific Silhouette Scores, the Adjusted Rand Index, and PCA visualization. Findings – K = 4 produced the highest overall Silhouette Score (0.4606) and the highest mean repeated-run ARI (0.9619), compared with K = 3 and K = 5. The selected solution comprised higher-score (n = 16), intermediate balanced (n = 24), mixed subject (n = 7), and lower-score profiles (n = 3). The first two principal components explained 87.87% of the variance. Research implications/limitations – The profiles are preliminary decision-support information and should not be used as formal classifications or as the sole basis for consequential educational decisions. Originality/value – The study provides a transparent and reproducible comparison of candidate solutions, including stability testing, inverse-transformed centroids, and cautious educational interpretation.]]>
