1 of 12

Forecasting Risk�Managing Food Hygiene Ratings during the COVID-19 Pandemic�

Sean Cunningham

Data Science Project Manager, Food Standards Agency

2 of 12

Background

The COVID-19 Pandemic has severely restricted workforces and strained hospital capacity

​

3 of 12

Project Goal

  • Develop a forecasting model to identify high-risk establishments to allow for reduced FSA staffing
  • Recommend for how long this model could be utilized while reduced staffing prevents 100% inspections

4 of 12

Main Findings

  • Forecasting Model:
    • Capable of supporting greater than a 50% reduction in workforce with no reduction in rate of identifying unsafe establishments

​

  • Model Longevity:
    • Model can be utilized for up to 9 months with decreased workforce with no reduction in rate of identifying unsafe establishments

5 of 12

Model Details

  • Utilized a Threshold Model
    • Parameters: Borough and Establishment Type
    • Predictions of “unsatisfactory” are marked for in-person inspection�
  • Maximized Information Gain
    • Balanced 4 elements of success to avoid “casting too wide a net” while still surveying the London Region

​

6 of 12

Newham and Hounslow

Waltham Forest and Barking

Lambeth

7 of 12

Requests for Information

  • Age of Establishment: New restaurants may be at higher risk for unsatisfactory conditions
  • Percentage of Staff Furloughed: Establishments with low staffing may be at higher risk for cutting corners, leading to unsatisfactory conditions.

8 of 12

Citations

  • Original Data Set:
    • https://data.gov.uk/dataset/e945d1f5-72fe-4b08-8b58-8e68d83441b5/uk-food-hygiene-rating-data-london-food-standards-agency
  • Understanding Food Safety Inspections:
    • https://www.food.gov.uk/safety-hygiene/food-hygiene-rating-scheme
  • Food Standards Agency Diversity Profile, 2018:
    • https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/932427/Food_Standards_Agency_Diversity_Report_2018.pdf
  • New York Times- United Kingdom COVID Map and Case Count:
    • https://www.nytimes.com/interactive/2020/world/europe/united-kingdom-coronavirus-cases.html

9 of 12

The Food Standards Agency, UK

10 of 12

Performance Statistics

  • Using a sample of >6000 establishments in the London Area, the model identifies unsatisfactory establishments more than twice as often as random inspections would identify.

​

  • This forecast was further tested on 12000 additional establishments in the London Area with less than a 10% decrease in effectiveness.

11 of 12

Modification of the Data Set

  • Deleted establishments without a rating
  • Deleted establishments without a latitude and/or longitude
  • Modified text to numbers
  • Removed 2/3 of unused columns (emails, city code etc.)
  • Reserved 2/3 of data for testing model
  • 1/3 of data used for forecasting

12 of 12