<![CDATA[Educational datasets increasingly contain multidimensional indicators describing school resources, participation, learning-related conditions, and institutional characteristics. When these indicators are analyzed simultaneously, direct clustering can be affected by dimensionality and correlated variables. This study develops a data-driven framework for identifying and assessing school phenotypes from multidimensional educational indicators using principal component analysis (PCA), clustering, and repeated-subsampling stability assessment. The analysis used the 2025 Rapor Publik Asesmen Nasional dataset for teachers in SD/MI/equivalent schools, comprising 283,948 observations and 92 variables. Twelve variables with complete missingness were removed, leaving 80 variables. A set of 33 core features was subsequently constructed, after which the administrative variable kd_kokab was excluded from the final analytical feature set, resulting in 32 features. Following imputation and scaling, PCA reduced the 32-dimensional feature space to seven components explaining 59.5413% of the total variance. K-means clustering applied to the seven-dimensional representation produced two clusters containing 114,399 (40.29%) and 169,549 (59.71%) observations, respectively. The final clustering yielded a Silhouette Score of 0.268125 and a Calinski–Harabasz index of 11,946.661. Stability assessment based on ten repeated samples of 30,000 observations produced a mean adjusted Rand index (ARI) of 0.989183 (SD = 0.006497), with ARI values ranging from 0.979404 to 0.997731. The findings establish a reproducible data-driven structure of two school groups and demonstrate high assignment stability under repeated subsampling. The framework provides an empirical basis for subsequent characterization of the indicators that differentiate the identified school profiles.]]>
