Machine Learning Models Predict Obesity Risk

By Muhammad OsamaReviewed by Susha Cheriyedath, M.Sc.Aug 6 2024

In a recent article published in the Journal of Personalized Medicine, researchers proposed a new technique to predict overweight and obesity risk by combining machine learning (ML) techniques. This algorithm considers sociodemographic, lifestyle, and health factors. The goal was to compare different models and identify the most important factors for weight gain.

Background

Obesity is a major global health issue, linked to increased risks of chronic diseases like diabetes, heart disease, and cancer. Tackling obesity requires looking at both personal factors (like genetics, age, and metabolism) and environmental factors (such as food access, socioeconomic status, and living conditions). Behavioral aspects, including diet, exercise, and lifestyle choices, are also crucial.

Artificial intelligence (AI), particularly ML, offers a promising way to predict and prevent obesity. AI can analyze large and complex datasets, like electronic health records, fitness tracker data, and dietary logs, to identify important patterns and variables. However, relying solely on ML might not be enough for effective obesity prediction. Therefore, combining AI with other methods and data sources is needed for a more accurate and comprehensive approach to obesity prevention and management.

About the Research

In this paper, the authors focused on designing and developing an interpretable model for predicting obesity or overweight risk using a combination of different ML algorithms. They used data from 1179 residents in Madrid, who participated in the EPIRCE (Epidemiology of Respiratory Chronic Diseases in Spain) study. The data included 38 variables related to lifestyle, sociodemographic, and health aspects, such as age, sex, education, occupation, physical activity, smoking, alcohol consumption, diet, blood pressure, cholesterol, and body mass index (BMI).

The researchers trained and tested nine classical ML techniques, such as decision tree, k-nearest neighbors, support vector machine, and neural network. They compared their performance with the cascade classifier model, which combines gradient boosting, random forest, and logistic regression models. Then, they used metrics such as accuracy, precision, and recall to evaluate the predictive results for overweight/obesity. They also used the shapely additive explanation (SHAP) technique to identify the factors/variables with the highest impact on weight gain.

Research Findings

The outcomes showed that the cascade classifier model provided the best predictive results for obesity and overweight, achieving an accuracy of 79%, recall of 89%, and precision of 84%. The SHAP analysis identified sex, age, education level, wine consumption, smoking habits, occupation, and adherence to the Mediterranean diet as the most influential factors. Specifically, older age, male sex, lower education, manual jobs, smoking, higher wine consumption, and lower adherence to the Mediterranean diet were linked to a higher risk of obesity. Some factors, like physical activity, fruit consumption, and olive oil intake, had a lower impact than expected.

Additionally, the authors found that location, marital status, mental and emotional disorders, diagnosed hypertension, and consumption of sweet drinks, fatty/oily foods, grilled foods, preserved foods, seasoning powders, soft drinks, and alcoholic beverages were also significant predictors of obesity. Furthermore, they compared their cascade classifier model with other studies using different datasets and ML techniques and found that their model achieved higher accuracy and recall than most previous studies.

Applications

This research has important implications for personalized medicine and public health. It offers a novel and accurate prediction algorithm for obesity risk, which can help identify individuals at high risk and provide personalized recommendations and interventions. The transparent AI system helps health professionals and researchers in understanding the complex relationships between obesity and various factors, enabling the design of more effective and tailored strategies. The study also showcases the potential of AI and ML to improve health outcomes and advance personalized medicine.

Conclusion

In summary, the novel approach based on the combination of different ML techniques proved effective for improving the accuracy of predicting risks of obesity or overweight as compared to a single ML algorithm. It effectively showed the several factors that impact the prediction of obesity the most.

Furthermore, the researchers recognized the limitations of their approach and suggested future work should explore other ML techniques, like deep learning and reinforcement learning, to improve the algorithm's predictive power. They also recommended validating the algorithm in different populations and settings and incorporating additional variables, such as genetic and epigenetic factors, for a more comprehensive and personalized prediction of obesity risk. They emphasized the importance of testing the model's generalizability and applicability across various populations and integrating other data sources and variables.