Clustering Provincial Democratic Performance in Indonesia Using Machine Learning

This study examines the regional variation of the Indonesian Democracy Index (IDI) across 34 provinces from 2021 to2024 in the context of global democratic backsliding. Understanding spatial disparities in democracy is crucial forpromoting inclusive and evidence-based governance. Using the Random Forest Clustering (RFC) method, provinces weregrouped based on indicators of civil liberties, political institutions, and community participation. Model selection wasoptimized by the Bayesian Information Criterion (BIC), with performance evaluated through R², silhouette score, and theCalinski–Harabasz index. The optimal model produced four clusters (R² = 0.806; silhouette = 0.310), revealing significantspatial diversity in democratic quality. Cluster 1 represents provinces with moderate stability and high heterogeneity;Cluster 2 includes those with limited democracy and low participation; Cluster 3 demonstrates consistent and balanceddemocratic performance; and Cluster 4 contains provinces with cohesive and mature democratic institutions. These resultshighlight interregional inequalities in democracy linked to institutional capacity and civic engagement. At the national level,the findings guide more targeted policy interventions for democratic strengthening. At the global level, this researchcontributes to understanding democratic resilience in developing countries and demonstrates the analytical value of machinelearning for comparative studies of governance and political systems