Predicting nationwide obesity from food sales using machine learning.

Area of Science:

• Public Health
• Epidemiology
• Data Science

Background:

• The global obesity epidemic poses a significant public health challenge.
• Nationwide obesity surveys are expensive and infrequent.
• Alternative, cost-effective methods are needed to monitor obesity prevalence.

Purpose of the Study:

• To develop and validate a methodology for predicting country-level obesity prevalence.
• To assess the feasibility of using national food sales data for obesity prediction.
• To identify specific food and beverage categories most predictive of obesity.

Main Methods:

• Utilized machine learning algorithms (support vector machines, random forests, extreme gradient boosting) for nonlinear regression.
• Employed country-level food sales data and obesity prevalence data from 79 countries.
• Validated the prediction model based on absolute error and satisfactory prediction proportions.

Main Results:

• The study successfully predicted country-level obesity prevalence using food sales data.
• Baked goods and flours emerged as the most significant food category for predicting obesity.
• Cheese and carbonated drinks were also identified as relevant predictors.

Conclusions:

• National food sales data offers a cost-effective alternative for monitoring obesity trends.
• Machine learning models can accurately forecast obesity prevalence using aggregated food purchase data.
• Targeting specific food categories like baked goods, flours, cheese, and carbonated drinks can inform public health interventions.