Data collection and modelling for food business compliance indicators

The purpose of this project was to collect a detailed level of data on the food related activities of a large sample of food establishments and to use this data to develop models to forecast how compliant establishments with particular characteristics are likely to be with food safety law. The focus for this work was to establish the framework for prediction models for new establishments who have yet to have a food safety inspection and who do not have any enforcement history. The model(s) are intended to inform a new food business segmentation model which will determine, based on risk, how food businesses are regulated in the future. The project has the following two core elements: • Data collection • Analysis of data and development of prediction models
Data collection The first objective of the project was to gather a large data set relating to the business activities of a range of food establishments, including assessments of legal compliance with food law by the relevant inspecting Local Authority (LA). This data is routinely gathered by LAs through their 9 official control work, such as inspections. The data is held by LAs, but not in a common or standard format. For the majority of local authorities this detailed level of information was expected to be held in paper reports, scanned documents or on databases. The data required included fields relating to: • Establishment (Name, address, premises type, opening hours, seasonality, cuisine type); • Food business operator (FBO) (franchise, sales activity);
• Food activities (food handling activities, food types handled, number of staff, method of processing, Food safety management system (FSMS) details, assurance, training, water supply, waste disposal contract, pest control contract); • Import/Export of food;
• Food hygiene inspection results; • Food standards inspection results. The full list is available in Appendix E. Development of model The second objective of the project was to use this data to develop a concept model for a risk engine that will segment new food establishments so that the most effective and proportionate initial intervention can be determined. The requirement was to predict the following outcomes in order of priority: • Seriously Non-Compliant (Food Hygiene Rating Scheme2 (FHRS) equivalent ratings of 0, 1) • Not Broadly Compliant (FHRS equivalent rating of 0, 1, 2) • Fully compliant premises (FHRS equivalent rating of 5) And to develop separate sub models such as for restaurants & catering