Hybrid Sampling and Ensemble Learning for Food Safety Sampling Inspection Classification.

Area of Science:

• Food safety
• Machine Learning
• Data Science

Background:

• Extreme class imbalance in food safety sampling biases ML models against detecting unqualified samples.
• Conventional oversampling methods struggle with complex food inspection data, leading to poor detection.
• Addressing nonlinear features, complex distributions, and multiclass scenarios is crucial for accurate food safety analysis.

Purpose of the Study:

• To develop a novel hybrid sampling algorithm to overcome limitations of conventional methods in food safety inspection.
• To enhance the detection of unqualified food samples by mitigating class imbalance and improving model robustness.
• To introduce an intelligent framework for food safety regulation using advanced ML techniques.

Main Methods:

• Proposed LOF-KNN-CSENN: a hybrid algorithm combining SMOTE and ENN with LOF for noise filtering and KNN for boundary preservation.
• Implemented a stacking ensemble learning framework with six tree-based models and Logistic Regression (LR) as a meta-model.
• Utilized Shapley Additive Explanations (SHAP) for identifying key risk factors in food safety.

Main Results:

• LOF-KNN-CSENN effectively suppressed noisy sample synthesis and balanced data distribution in food inspection datasets.
• The integrated stacking ensemble model achieved higher precision (0.4-5.6%) and F1-score (0.8-30.7%) compared to single models.
• SHAP analysis identified production address, sampling stage, and location as critical risk factors for targeted food safety supervision.

Conclusions:

• The proposed LOF-KNN-CSENN algorithm and stacking ensemble framework offer a robust solution for intelligent food safety regulation.
• This approach significantly enhances the detection of unqualified samples in multicategory food inspection by addressing class imbalance.
• The findings support targeted supervision strategies by highlighting key risk factors, contributing to improved food safety outcomes.