<![CDATA[Abstrak : Penetapan majelis hakim secara manual di Pengadilan Agama Sragen sering kali memerlukan waktu hingga tiga hari dan berisiko menimbulkan ketidakefisienan, terutama ketika jumlah perkara tinggi dan ketua pengadilan berhalangan hadir. Penelitian ini bertujuan mengembangkan sistem rekomendasi susunan majelis hakim perkara tingkat pertama dengan metode Support Vector Machine (SVM) untuk memberikan rekomendasi yang lebih cepat dan efisien kepada Ketua Pengadilan Agama Sragen. Sistem dibangun menggunakan pendekatan hybrid, menggabungkan model klasifikasi SVM untuk perkara umum dan rule-based untuk perkara khusus seperti ekonomi syariah dan dispensasi kawin. Dataset yang digunakan terdiri atas 1.428 data perkara dan delapan profil hakim, yang diproses melalui tahap pembersihan teks, ekstraksi kata kunci, transformasi fitur dengan TF-IDF dan one-hot encoding, serta balancing data menggunakan SMOTE. Model dilatih dengan kernel RBF dan divalidasi melalui 5-fold cross-validation. Hasil pengujian menunjukkan model ketua majelis mencapai F1-score 64%, akurasi 63%, presisi 66%, dan recall 63%. Sementara model anggota 2 memperoleh F1-score 56%, dan anggota 1 sebesar 36%. Confusion matrix memperlihatkan bahwa sistem dapat mengenali pola penugasan hakim dominan namun masih kesulitan menangani kelas minoritas. Sistem ini menunjukkan potensi sebagai alat bantu rekomendasi yang mampu meningkatkan efisiensi, akurasi, dan konsistensi dalam penetapan majelis hakim.=================================================Abstract :Manual assignment of judicial panels at the Sragen Religious Court often takes up to three days and poses a risk of inefficiency, especially when the case load is high and the chief judge is unavailable. This study aims to develop a recommendation system for composing first-instance judicial panels using the Support Vector Machine (SVM) method, in order to provide faster and more efficient recommendations to the Chief Judge of the Sragen Religious Court. The system is built using a hybrid approach, combining an SVM classification model for general cases and a rule-based approach for special cases such as Islamic finance and marriage dispensation. The dataset used consists of 1,428 case records and eight judge profiles, processed through text cleaning, keyword extraction, feature transformation using TF-IDF and one-hot encoding, and class balancing using SMOTE. The model is trained with the RBF kernel and validated using 5-fold cross-validation. Evaluation results show that the model for the chief judge achieved an F1-score of 64%, accuracy of 63%, precision of 66%, and recall of 63%. Meanwhile, the model for member 2 reached an F1-score of 56%, and member 1 only 36%. The confusion matrix reveals that the system can recognize dominant judge assignment patterns, although it still struggles with minority classes. This system demonstrates strong potential as a recommendation tool that can enhance efficiency, accuracy, and consistency in judicial panel assignments.]]>
