How a ‘recipe dataset’ could help tackle obesity and diabetes

This article is brought to you thanks to the collaboration of The European Sting with the World Economic Forum.

Author: Johnny Wood, Senior Writer, Formative Content

• Machine learning is being used by a US healthy eating programme to understand people’s food preferences, health condition and dietary aims.
• The system suggests recipes tailored to a person’s tastes and health needs.
• The AI-enabled programme aims to help diabetics, people with obesity, or those with heart conditions live longer.

Low-fat. High-fibre. Calorie-loaded. Sugar-free. Maintaining a healthy diet that’s right for your individual body and lifestyle isn’t easy – unless you add some AI into the recipe.

A machine-learning algorithm that monitors food preferences and makes nutritious recipe suggestions tailored to each individual’s needs has been devised by scientists at Rensselaer Polytechnic Institute and IBM Research, both in New York. The programme notes personal likes and dislikes, allergies and other factors to guide healthy eating.

The system’s name is rather a mouthful – pFoodReQ – but it could help inform daily food choices and provide eating prompts for diabetics, people with heart conditions or those pursuing a healthier diet.

New-trition

The new food programme, part of a broader project called Health Empowerment Analytics Learning & Semantics (HEALS), employs AI deep learning to refine recipe suggestions and distinguish if a meal or a snack is required at different times of the day.

“Our work focuses on personalized food recommendations,” Mohammed J Zaki, a researcher on the project, told Tech Xplore.

People enrolled on the HEALS programme record their food preferences, medical condition and dietary goals and keep a log of their food intake, which provides material for the system to learn over time about their individual eating habits.

The system learns to distinguish the type of meal required, when it should be consumed and suggests nutritious recipes – from what Zaki calls a “recipe dataset” – aimed at satisfying individual tastes, while also promoting good health and accounting for allergies or other food constraints.

Healthy habits

For diabetics, healthy recipe suggestions that control the intake of carbohydrates, proteins and other food groups could help keep sugar levels stable, along with reminders to eat regular meals.

The International Diabetes Federation (IDF) estimates there were 451 million people living with diabetes worldwide in 2017, a number projected to increase to more than 693 million by 2045 without preventative action. Diabetes is among the top 10 causes of deaths among the global population, according to a study in Nature.

While incidents of diabetes have increased across all world regions, rates are predicted to grow fastest in Africa, the Middle East and Southeast Asia in the coming years. By 2045, the number of diabetics in Africa aged between 20 and 79 years, is forecast to exceed 2019 levels by 143%, according to Statista figures.

As with diabetes, incidents of other diet-related conditions like obesity are also increasing. Between the turn of this century and 2018, age-adjusted obesity rates increased from around 30% to more than 42% of the US population. The number of severely obese people in America has almost doubled over the same period.

One in five global deaths is linked to poor diet, which is equivalent to 11 million premature deaths, according to the Global Burden of Disease study published in The Lancet. But the situation could look very different as machine learning takes the guesswork out of the food choices we make.