High-Efficiency Machine Learning Method for Identifying Foodborne Disease Outbreaks and Confounding Factors

Area of Science:

• Food safety
• Epidemiology
• Machine learning

Background:

• The China National Center for Food Safety Risk Assessment (CFSA) relies on the Foodborne Disease Monitoring and Reporting System (FDMRS) for national foodborne disease surveillance.
• Existing FDMRS data faces challenges with underreporting and erroneous entries, increasing epidemic investigation costs.

Purpose of the Study:

• To develop and evaluate a machine learning model for accurately identifying suspected foodborne disease outbreaks from FDMRS data.
• To improve the efficiency and reduce the cost of foodborne disease outbreak investigations.

Main Methods:

• Machine learning models were employed to analyze FDMRS data.
• The eXtreme Gradient Boosting (XGBoost) model was selected for its superior performance.
• Recall rate and F1-score were used as key evaluation metrics.
• SHapley Additive exPlanation (SHAP) values were utilized for feature importance analysis.

Main Results:

• The XGBoost model demonstrated the highest performance with a recall rate of 0.9699 and an F1-score of 0.9582.
• Model validation confirmed its accuracy in identifying real foodborne disease outbreaks.
• Feature importance analysis highlighted the health status of similarly exposed individuals as a critical factor (weight of 0.65).

Conclusions:

• The developed machine learning model accurately detects foodborne disease outbreaks, significantly reducing the manual workload for medical staff.
• The study identified key confounding factors in foodborne disease outbreaks, emphasizing the importance of temporal and spatial case similarity alongside individual health status.