Skripsi
OPTIMASI PARAMETER SUPPORT VECTOR MACHINE (SVM) BERBASIS GRID SEARCH DENGAN SMOTE UNTUK KLASIFIKASI RISIKO KREDIT NASABAH
The banking industry faces significant challenges in credit risk management due to the subjectivity of manual analysis. This study aims to develop an objective credit risk classification model using Support Vector Machine (SVM) optimized with Grid Search (GS) and Synthetic Minority Over-sampling Technique (SMOTE) on the German Credit Dataset. Recall is designated as the primary evaluation metric to minimize financial losses from false negatives. Testing results show that the Baseline SVM model only achieves a recall of 0.48. However, the integrated SVM + SMOTE + GS model significantly improves the recall to 0.82 and the ROC-AUC to 0.80. This hybrid approach slashes the number of false negatives from 31 to just 11 customers representing a 64.5% improvement in detection of high-risk customers. This study demonstrates that the combination of synthetic data balancing and systematic hyperparameter optimization enhances the model's recall sensitivity in detecting high-risk customers for banking decision support systems.