Toward an Early Warning System: Feature Selection Mechanism for Food Fraud from Fishing Activities

Food fraud presents a considerable challenge to consumer safety, economic stability, and the integrity of global food supply chains. This study introduces a Feature Selection Mechanism for Fraud (FSMF) designed to proactively monitor and identify fraudulent activities within food supply chains, with a particular emphasis on the fish supply chain, and more specifically focusing on fishing stage. The system integrates IoT sensor data, environmental factors, and supply chain records, employing machine learning techniques for anomaly detection and feature extraction. The key methodologies used in this work include, Isolation Forest, Autoencoders, ARIMA, in addition to multi-modal data fusion and feature importance selection approaches. To illustrate the proposed system’s effectiveness, a case study focusing on the Norwegian whitefish, utilizing Random Forest and XGBoost for vessel classification. Explainable AI techniques, such as SHAP analysis, are implemented to examine the impact of environmental and behavioral features on classification accuracy. The findings reveal a significant enhancement in fraud detection, achieving high precision and recall in differentiating fishing activities from other maritime operations.