🏥 XAI-based Childhood Weight Management System

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

An innovative framework leveraging wearable devices and explainable AI to address childhood obesity through personalized predictions and interpretable insights.

🌟 Key Features

• 📊 IoT Data Integration: Seamless collection from Samsung Galaxy Fit 2 and smartphones
• 🤖 Hybrid AI Model: TabNet + XGBoost architecture for superior performance
• 🔍 Explainable AI: SHAP and TabNet mask mechanisms for interpretability
• ⚖️ Synthetic Data Generation: Advanced GAN-based techniques for class balance
• 👶 Child-Focused: Specifically designed for elementary school children
• 📱 Real-time Monitoring: Continuous lifestyle pattern analysis

🏗️ System Architecture

'''

📋 Requirements

python>=3.8 tensorflow>=2.0 xgboost>=1.5.0 pytorch-tabnet>=3.1.1 shap>=0.41.0 numpy>=1.21.0 pandas>=1.3.0 scikit-learn>=1.0.0 nbsynthetic>=0.1.0

📊 Dataset Information

Data Collection

• School A (Seoul): 362 elementary students, 6 months, 44,226 records
• School B (Jeju): 82 elementary students, 8 weeks, 3,343 records
• Devices: Samsung Galaxy Fit 2, WUD! app, Samsung Health

Features Collected

Feature	Source	Description
Height/Weight	Manual entry	Regular updates by participants/parents
Calorie Intake	Food photos/manual	Based on National Food Nutrition database
Step Count	Smartwatch	Daily physical activity
Sleep Duration	Smartwatch/smartphone	Sleep pattern analysis
Burned Calories	Smartwatch	Energy expenditure

Target Labels

• Label 1: Weight Loss (1.15%)
• Label 2: Weight Maintenance (96.15%)
• Label 3: Weight Gain (2.7%)

🧠 Model Architecture

Hybrid TabNet-XGBoost Model

📈 Performance Results

Model Performance (Test Dataset)

Class-wise Performance

• Weight Loss: 96.73% accuracy
• Weight Maintenance: 99.55% accuracy
• Weight Gain: 92.68% accuracy

External Validation (School B)

• Accuracy: 85.2%
• F1-Score: 81.4%