<![CDATA[The high risk of credit default and class imbalance in customer data pose major challenges in developing accurate credit scoring systems. This condition causes predictive models to be biased toward the majority class, thereby reducing the ability to detect high-risk borrowers. This study develops a credit scoring model for imbalanced data using the Synthetic Minority Oversampling Technique (SMOTE) and cost-sensitive Random Forest with hyperparameter optimization via GridSearchCV. The dataset consists of 32,581 customer records. Experimental results show that the best configuration with n_estimators = 200 achieves a cross-validation F1-score of 0.813750. On the test data, the model attains an accuracy of 0.927267, precision of 0.911458, recall of 0.738397, and an F1-score of 0.815851, indicating improved and more balanced detection of high-risk borrowers.Keywords: Random Forest; SMOTE; Cost-sensitive learning; GridSearchCVÂ AbstrakTingginya risiko gagal bayar kredit dan ketidakseimbangan kelas pada data nasabah menjadi tantangan utama dalam pengembangan sistem credit scoring yang akurat. Kondisi ini menyebabkan model prediksi cenderung bias terhadap kelas mayoritas sehingga kemampuan deteksi debitur berisiko menjadi kurang optimal. Penelitian ini mengembangkan model credit scoring pada data tidak seimbang menggunakan Synthetic Minority Oversampling Technique (SMOTE) dan Cost-Sensitive Random Forest dengan optimasi hyperparameter GridSearchCV. Dataset yang digunakan berjumlah 32.581 data nasabah. Hasil pengujian menunjukkan konfigurasi terbaik dengan n_estimators = 200 menghasilkan F1-score validasi silang sebesar 0,813750. Pada data uji, model mencapai akurasi 0,927267, precision 0,911458, recall 0,738397, dan F1-score 0,815851, yang menunjukkan peningkatan kemampuan deteksi debitur berisiko secara lebih seimbang.Kata kunci: Random Forest; SMOTE; Cost-sensitive learning; GridSearchCV.]]>
Copyrights © 2026
