Anomaly Score-Based Risk Early Warning System for Rapidly Controlling Food Safety Risk

Area of Science:

• Food Science
• Artificial Intelligence
• Risk Management

Background:

• Food safety is a critical global concern requiring robust early warning and risk control systems.
• Effective management of food safety relies on accurate detection and classification of product quality.

Purpose of the Study:

• To develop an innovative anomaly score-based risk early warning system (ASRWS) for food products.
• To classify products as qualified or unqualified using an unsupervised auto-encoder (AE) model based on reconstruction errors.

Main Methods:

• An unsupervised auto-encoder (AE) model was employed for anomaly detection in food product testing data.
• The system utilizes reconstruction errors to classify products and early warning thresholds for risk analysis.
• A hybrid approach combining AI predictions with expert risk revision was implemented.

Main Results:

• The AE model demonstrated high prediction accuracy (0.9954) and fault detection rate (0.9024) on dairy product data.
• The system achieved rapid analysis within 0.54 seconds.
• Expert revision enhanced the reliability of AI-driven food safety predictions.

Conclusions:

• The proposed ASRWS offers a fast, cost-effective solution for food safety early warning using detection data.
• AI integration improves the efficiency of food safety expert panels.
• The study supports market supervision departments in proactively controlling food safety risks.