Leveraging Machine Learning to Assess Market-Level Food Safety and Zoonotic Disease Risks in China

Project Description & Overview

In recent years, China has faced significant challenges with zoonotic diseases originating from its wholesale and wet markets, highlighting the critical need to balance food safety regulations with market sustainability. This project aims to develop a data-driven approach using machine learning to evaluate and mitigate these risks effectively.

Objectives

The primary objective of this project is to assess the correlation between food safety risks (specifically, food adulteration) in wholesale markets (WSMs) and wet markets (WMs) across China and the incidence of zoonotic diseases. By analyzing extensive datasets derived from regulatory food safety tests and zoonotic disease outbreaks, the project seeks to:

• Develop Market-Level Risk Scores: Utilize machine learning algorithms to create food adulteration risk scores for individual markets based on historical regulatory data from China's Administration for Market Regulation (AMR).
• Correlate Risk Scores with Zoonotic Diseases: Investigate and establish statistical correlations between these market-level risk scores and the occurrence of zoonotic disease outbreaks using epidemiological data.
• Identify High-Risk Markets: Identify and prioritize markets with elevated food adulteration risks, potentially linked to zoonotic outbreaks, to guide targeted regulatory interventions.

Methodology

The project employs a comprehensive methodology integrating machine learning techniques and epidemiological analyses:

• Data Collection and Preparation: Aggregate and preprocess a vast dataset of over 4 million food safety test records conducted between 2014 and 2020 by AMRs at various levels (state, provincial, prefecture).
• Machine Learning for Market Identification: Develop novel clustering algorithms tailored for Chinese text to associate food safety test records with specific markets, overcoming challenges in inconsistent market naming conventions.
• Risk Score Calculation: Compute market-level food adulteration risk scores based on failure rates from AMR tests, focusing on markets handling high volumes of animal products.
• Statistical Analysis: Apply regression models to assess the relationship between these risk scores and zoonotic disease outbreaks, controlling for regional factors like livestock production and market density.

AMR Pipeline Process

This image illustrates the flow chart diagram of the AMR Pipeline process:

• Step I: AMR Food Adulteration Database Integration
• Step II: Identify AMR Database Test Records of Animal Products at Markets
• Step III: Unsupervised Clustering to Connect Records to Specific Markets
• Step IV: Market Risk Scores (Failure Rates)
• Step V: Province Risk Scores (Number of AMR Tests at Markets in the Top 20% of Risk Scores)
• Step VI: Province Analysis (Association of Province Risk Scores with Zoonotic Flu Isolates in Humans)

Results and Impact

The project's outcomes aim to provide actionable insights and tools for policymakers, regulators, and public health officials:

• Decision Support Tool: Develop an operational tool for visualizing market risk scores and associated food safety test results, aiding regulators in prioritizing inspections and interventions.
• Policy Recommendations: Inform evidence-based policies for targeted regulatory interventions that enhance food safety and mitigate zoonotic disease risks while preserving market operations vital to China's food supply chain.
• Contribution to Public Health: Advance understanding of the systemic factors contributing to zoonotic disease transmission in market settings, facilitating proactive public health measures.

Sample Subset Table of AMR Dataset

File: Zhejiang_Zhejiang_msb_20220730/3.食品抽检合格-20201111.xls.

Sample Subset Table of AMR Dataset (English Translations)

File: Zhejiang_Zhejiang_msb_20220730/3.Food sampling qualified-20201111.xls.