<![CDATA[Educational inequality across Indonesia’s 38 provinces remains a challenge to equitable development. This study identifies anomalous provincial educational profiles by comparing Isolation Forest, Local Outlier Factor, and One-Class SVM across 50 parameter configurations in five analytical categories. The framework integrates ensemble majority voting, SHAP-based attribution, and leave-one-out robustness testing. Under the Silhouette-based selection criterion, Isolation Forest with contamination 0.05 ranked highest in four categories, with Silhouette Scores of 0.47-0.60 and Stability Scores of 1.0. However, sensitivity analysis showed that this criterion tends to favor configurations detecting fewer anomalies, so IF (0.05) should not be considered unambiguously superior. Highland Papua was consistently identified as anomalous across all categories. SHAP highlighted socioeconomic indicators, particularly rural child-labor participation, as important contributors to anomaly scores. Given the single-year design, 38 observations, and high feature-to-sample ratio, findings should be interpreted as exploratory descriptive mapping rather than causal or broadly generalizable inference.]]>
Copyrights © 2026
