A Smart System to Empower Healthy Food Choices - Machine Learning Component, 2020-2021

Holford, Dawn (2023). A Smart System to Empower Healthy Food Choices - Machine Learning Component, 2020-2021. [Data Collection]. Colchester, Essex: UK Data Service. 10.5255/UKDA-SN-855340

Proper nutrition and healthy diets are a key aspect of health, which mandatory food labelling in the UK tries to address by empowering people with the information to help them make healthier choices. The format of this information (e.g., verbal quantifiers like 'low fat' or numerical quantifiers like '5% fat') affects whether people can easily understand and use food labels. Examining how people's judgements and decisions with respect to food differ depending on food label format therefore has wide-reaching impact for health policy decisions, consumer behaviour, and food industry practice. This project will use computational methods to identify different strategies people use to decide what foods are healthiest (e.g., less fat, or less sugar, etc.) I will evaluate which strategies produce the healthiest choices, use these insights to inform policy and conduct knowledge exchange with my industry partner.

The project will consolidate my PhD, which investigated differences in people's decision-making strategies when using verbal and numerical quantifiers on food labels. Using a mixture of behavioural tasks, surveys, and eye-tracking methodology, I identified that different ways of presenting quantities can lead to people relying on different pieces of information to judge food. I intend to extend this research and maximise its impact in four ways.

First, I will apply new and advanced statistical modelling to my research. To classify and predict food choice strategies in my data, I will learn two modelling techniques: multinomial processing trees, a probability-based method to classify choices, and machine learning, which makes predictions based on patterns in data. For example, I would expect the models to identify cues on food labelling that predict the choices people will make. Using the results of these analyses, I will submit a planned research protocol (a 'Registered Report') to test my model on real-life products. Registered Reports receive peer review prior to data collection, so submitting it during the Fellowship supports my future academic research beyond the Fellowship.

Second, I will extend the impact of my work through knowledge exchange with the start-up company Keep Fit Eat Fit Wellbeing Ltd (KFEF). As part of a holistic wellness package, KFEF produces healthy eating advice and recipes with nutritional information for their clients. My research will inform the design of their content for clients. In turn, working with them gives me access to usage metrics from their customer portal that I will analyse to determine if the communication formats are effective. These real-world data will reinforce the lab studies from my PhD and help KFEF improve their product offering.

Third, I will disseminate my research findings to academic and non-academic audiences. For academic audiences, I will produce three new journal articles and present my work at one local and one international academic conference. I will also engage with non-academic audiences through preparing press releases, submitting a policy brief to present at the All-Party Parliamentary Food and Health Forum, and attending a Westminster Food and Nutrition Forum conference. Engaging with policy-makers through these channels will help me lobby for positive change to food labelling guidelines.

Finally, I will prepare a proposal for funding from the Wellcome Trust to create and test a technological system that supports informed food choices. This future proposal will be informed by my PhD data, computational modelling research, and collaborations with: industry (Keep Fit Eat Fit), experts in shaping behavioural policy (at the University of Reading), and experts in technological health interventions (at the University of Konstanz). Ultimately, my research seeks to improve the food choice environment for consumers and empower them to make informed, healthy choices.

Data description (abstract)

Dataset and associated material used for the machine learning analysis of food choices. The dataset was obtained from an experiment with 154 participants who made 30 choices of the healthiest food within a choice array of 6 options, given nutritional label data. This is a secondary data analysis and the original data collection was not funded by the grant. The dataset contains 4260 observations (response trials) and 51 variables.

Data creators: