<![CDATA[Abstrak - Pemanfaatan machine learning pada data administratif berpotensi mendukung analisis berbasis data, namun kinerjanya sangat dipengaruhi oleh karakteristik data yang digunakan. Penelitian ini bertujuan untuk mengevaluasi kinerja beberapa algoritma machine learning dalam mengklasifikasikan tingkat pendidikan warga binaan berdasarkan data administratif non-identitas. Algoritma yang digunakan meliputi Support Vector Machine (SVM), Random Forest, dan XGBoost. Data yang digunakan berasal dari dokumen administrasi internal lembaga pemasyarakatan dan diproses melalui tahapan pra-pemrosesan, pembagian data latih dan data uji, serta evaluasi model menggunakan metrik akurasi, presisi, recall, dan F1-score. Hasil penelitian menunjukkan bahwa SVM (RBF) menghasilkan nilai akurasi tertinggi, sementara XGBoost memiliki performa yang relatif lebih seimbang pada metrik presisi, recall, dan F1-score. Namun demikian, seluruh algoritma menunjukkan kinerja yang masih tergolong rendah. Temuan ini mengindikasikan bahwa keterbatasan data administratif, khususnya dominasi atribut kategorikal dan distribusi kelas yang tidak seimbang, memengaruhi kemampuan algoritma dalam mengenali pola secara optimal. Penelitian ini memberikan gambaran empiris mengenai efektivitas dan keterbatasan penerapan machine learning pada data administratif non-identitas.Kata kunci : Machine learning; Evaluasi kinerja; Data administratif; Klasifikasi; Warga binaan; Abstract - The use of machine learning on administrative data has the potential to support data-driven analysis, but its performance is greatly influenced by the characteristics of the data used. This study aims to evaluate the performance of several machine learning algorithms in classifying the educational level of inmates based on non-identifying administrative data. The algorithms used include Support Vector Machine (SVM), Random Forest, and XGBoost. The data used comes from internal administrative documents of correctional institutions and is processed through pre-processing, training and testing data division, and model evaluation using accuracy, precision, recall, and F1-score metrics. The results show that SVM (RBF) produced the highest accuracy, while XGBoost had relatively more balanced performance in terms of precision, recall, and F1-score metrics. However, all algorithms showed relatively low performance. These findings indicate that limitations in administrative data, particularly the dominance of categorical attributes and unbalanced class distribution, affect the ability of algorithms to recognize patterns optimally. This study provides an empirical overview of the effectiveness and limitations of applying machine learning to non-identity administrative data.Keywords: Machine learning; Performance evaluation; Administrative data; Classification; Inmates;]]>
