EL V.2 Model for Predicting Food Safety Risks at Taiwan Border Using the Voting-Based Ensemble Method.

Area of Science:

• Food safety
• Public health
• Computational intelligence

Background:

• Border management is critical for ensuring imported food quality and safety.
• Taiwan implemented the first-generation ensemble learning prediction model (EL V.1) in 2020 for food border control.
• EL V.1 used five algorithms to assess imported food risk and guide sampling decisions.

Purpose of the Study:

• To develop a second-generation ensemble learning prediction model (EL V.2) for enhanced food safety at the border.
• To improve the detection rate of unqualified food products and increase model robustness.
• To compare the effectiveness of model-guided sampling with traditional random sampling.

Main Methods:

• Developed EL V.2 using seven algorithms, including Bagging-Gradient Boosting Machine and Bagging-Elastic Net.
• Utilized Elastic Net for characteristic risk factor selection.
• Employed Fβ to optimize sampling rates and a chi-square test for efficacy comparison.

Main Results:

• EL V.2 demonstrated superior predictive performance over EL V.1 and random sampling.
• The unqualified rates for model-predicted inspections (5.10%-6.36%) were significantly higher than random sampling (2.09%).
• EL V.2 showed improved detection of unqualified food cases and enhanced model robustness.

Conclusions:

• The second-generation ensemble learning model (EL V.2) significantly enhances the detection of unqualified imported food.
• Model-guided sampling is more effective than random sampling in identifying unsafe food products at the border.
• EL V.2 represents a significant advancement in food safety border management.