AI Framework Forecasts City-Level Food Safety Risks Using Sparse Data
A recent study presents a Transformer-based model aimed at predicting detailed food safety risks at the city level, tackling issues related to limited inspection resources and sparse sampling data. This research, available on arXiv, combines more than 11 million inspection records with demographic, economic, and environmental data sourced from the Statistical Yearbook. Utilizing a three-stage pretraining approach, it employs partial supervision via the Wilson interval and semi-supervised label refinement to make the most of historical data despite inadequate local samples. Tests conducted on 2022 data indicate that this method significantly surpasses existing baselines. Additionally, a field experiment with the Zhejiang Provincial Administration for Market Regulation showcased enhancements in practical applications.
Key facts
- The framework is Transformer-based and forecasts city-level food safety risks.
- It unifies over 11 million inspection records with supplemental demographic, economic, and environmental indicators.
- A three-stage pretraining design uses partial supervision from the Wilson interval and semi-supervised label refinement.
- Experimental evaluations on 2022 data show the approach outperforms baselines significantly.
- A field experiment was conducted with the Zhejiang Provincial Administration for Market Regulation.
- The study addresses challenges of limited inspection resources and sparse regional sampling data.
- The paper is available on arXiv with identifier 2608.01767.
Entities
Institutions
- Zhejiang Provincial Administration for Market Regulation
- arXiv
Locations
- Zhejiang
- China