<![CDATA[This study evaluates Deep Support Vector Data Description (Deep-SVDD) for anomaly detection in credit insurance claim submissions processed through host-to-host systems. The model addresses irregularities such as duplicate claims, inconsistent values, and delayed reporting by learning normal claim behavior in a latent space and applying calibrated thresholds. Using a dataset of 5,000 claims with mixed-type variables, Deep-SVDD achieved strong performance on the validation set, with high precision, recall, and ROC-AUC. Confusion matrix and Recall@K analyses confirmed low false alarms and effective anomaly ranking, capturing a substantial portion of anomalies among top-ranked claims. These results demonstrate Deep-SVDD’s potential as a scalable and efficient early detection layer, improving transparency and reliability in credit insurance claim verification.]]>
Copyrights © 2025
