Combination of Machine Learning Techniques to Predict Overweight/Obesity in Adults

Area of Science:

• Public Health
• Artificial Intelligence
• Biomedical Informatics

Background:

• Obesity is a growing public health concern.
• Artificial intelligence (AI) and machine learning (ML) offer promising tools for predicting and preventing obesity.
• Developing interpretable prediction algorithms is crucial for clinical application.

Purpose of the Study:

• To design an interpretable prediction algorithm for overweight/obesity risk.
• To evaluate the performance of a combined ML approach against individual ML techniques.
• To identify key factors influencing weight gain.

Main Methods:

• Collected 38 variables (sociodemographic, lifestyle, health) from 1179 Madrid residents.
• Trained and compared nine classical ML techniques and a combined ML model.
• Utilized Shapley Additive Explanation (SHAP) for variable impact analysis.

Main Results:

• The cascade classifier model (gradient boosting, random forest, logistic regression) achieved the highest accuracy (79%), precision (84%), and recall (89%).
• Key predictors for obesity included age, sex, academic level, profession, smoking, wine consumption, and Mediterranean diet adherence.
• The combined ML model significantly outperformed individual ML techniques.

Conclusions:

• A combination of ML techniques significantly improves the accuracy of overweight/obesity risk prediction compared to individual methods.
• The developed interpretable model can aid in identifying individuals at high risk for weight gain.
• This AI-driven approach holds potential for obesity prevention strategies.