Anomaly Detection based on GS-OCSVM Classification

This research aims to apply one-class support vector machine classifier (OCSVM) for anomaly detection and estimate the hyperparameters of OCSVM using the grid search method. The proposed grid search one-class support vector machine algorithm (GS-OCSVM) is then applied to the fraud detection problem. Data used in this study consists of German credit card and European cardholder credit card transactions which treat the fraud transactions as anomalies. In this study, we estimated the values of the hyperparameters y and v of OCSVM by considering the maximum of area under the curve (AVC). The results show that the GS-OCSVM can detect fraud better than the isolation forest as true negative rate is higher than isolation forest for both datasets.