Application of Explainable Artificial Intelligence for personalized childhood weight management using IoT data.

Area of Science:

• Pediatric Health
• Artificial Intelligence in Medicine
• Digital Health Technologies

Background:

• Childhood obesity is a significant global health issue linked to chronic diseases.
• Digital health and AI offer potential for analyzing health data but face challenges like data limitations and interpretability.
• Existing research requires robust methods for real-time data analysis and personalized intervention in childhood obesity.

Purpose of the Study:

• To develop and validate a comprehensive AI framework for childhood obesity research.
• To address data imbalance and model interpretability challenges in AI for childhood obesity.
• To enable personalized health guidance and targeted interventions for preventing childhood obesity.

Main Methods:

• Utilized wearable devices for real-time lifestyle data collection.
• Employed Wasserstein generative adversarial networks (WGANs) to manage data imbalance.
• Integrated explainable AI models: Tabular Attention Network (TabNet) and eXtreme Gradient Boosting (XGBoost) with SHapley Additive exPlanations (SHAP).

Main Results:

• Achieved 98.0% accuracy on the internal test dataset.
• Demonstrated 85.2% accuracy on an external validation dataset.
• Successfully identified and explained individual factors contributing to weight change.

Conclusions:

• The proposed AI framework effectively addresses key challenges in childhood obesity research.
• The framework provides accurate predictions and interpretable insights for personalized interventions.
• This approach supports targeted strategies for childhood obesity prevention and management.