Social Network Analysis (SNA) on Predicting Shipment Frauds in Consignment Goods

The issue of fraud in consignment goods is a growing concern in the import sector. With an annual volume exceeding 61 million items, limited manpower among customs officers, smuggling of prohibited items and tax evasion there is an urgent need for effective tools to detect fraud. However, the highly imbalanced and feature-limited nature of consignment goods datasets makes accurate fraud classification challenging. This study explores techniques to leverage the dataset's unique characteristics, employing Social Network Analysis (SNA) to uncover intricate connections within consignment goods data. Five models with inherent strategies for handling imbalanced data were evaluated for fraud classification. The models were tested on three feature sets, with the final one combining SNA with personal features. Among the models, XGBoost achieved score of 1.000 on F1, Accuracy, Precision, Recall and AUC-PR Score. SNA proved effective in enhancing fraud classification on consignment goods datasets.