Chronic Disease Risk Predictor 🏥 - Lōkahi Innovation

An advanced AI-powered healthcare analytics platform for predicting chronic disease risks and providing personalized medical recommendations.

🌟 Key Features

• Advanced Risk Prediction: Utilizes XGBoost model with multi-condition analysis
• Interactive Dashboard: Real-time analytics and population health insights
• AI-Powered Recommendations: Personalized health guidance using Llama 3.2 model
• Secure Data Management: MySQL integration for robust data storage
• Dynamic Visualizations: Interactive charts and metrics using Plotly
• Modern UI: Responsive design with Streamlit components

🔧 Technology Stack

Frontend

• Streamlit (v1.40.2)
• Plotly for interactive visualizations
• Custom CSS styling

Backend

• Python 3.8+
• MySQL database - CDRPredictor Database
• Advanced logging system

AI/ML Components

• XGBoost for risk prediction
• Together AI (Llama-3.2-3B-Instruct-Turbo) for recommendations
• Scikit-learn for data preprocessing

Data Processing

• Pandas (v2.2.3)
• NumPy (v2.1.3)
• Joblib (v1.4.2)

📊 Dashboard Features

• Real-time Analytics: Monitor patient risk levels and trends
• Population Health Metrics: Track key health indicators
• Condition Distribution: Analyze prevalence of chronic conditions
• Advanced Filtering: Customize views by demographics and conditions
• Interactive Charts: Dynamic visualization of health trends

🚀 Getting Started

Prerequisites

• Python 3.8+
• MySQL Server - CDRPredictor Database
• pip package manager

Installation

1. Clone the repository:

   git clone https://github.com/ajitonelsonn/chronic_disease_predictor.git
   cd chronic_disease_predictor

2. Set up a virtual environment:

   python -m venv venv
   source venv/bin/activate  # Linux/Mac
   # or
   venv\Scripts\activate     # Windows

3. Install dependencies:

   pip install -r requirements.txt

4. Configure Environment: Create .streamlit/secrets.toml:

   [api_keys]
   togetherapi = "your_together_api_key"
   
   [database]
   db_host = "your_db_host"
   db_username = "your_db_username"
   db_password = "your_db_password"
   db_name = "your_db_name"
   db_port = "your_db_port"

📁 Project Structure

chronic_disease_predictor/
├── .streamlit/
│   ├── config.toml
│   └── secrets.toml
├── components.py
├── pages/
│   ├── dashboard.py
├── database/
│   └── schema.sql
├── model/
│   ├── best_chronic_disease_model.joblib
│   ├── feature_scaler.joblib
│   └── label_encoder.joblib
├── utils.py
├── styles.py
├── database.py
├── model_utils.py
├── recommend.py
├── streamlit_app.py
└── requirements.txt

---

🤖 Model Development

Detailed documentation of our model development process can be found in our Jupyter Notebook or Create Model.

Data Processing

• Processed 450,000 patient records with 37.5M entries
• Integrated data from multiple sources:
  • Member demographics
  • Enrollment history
  • Service records
  • Provider information

Model Selection Process

We evaluated three different models:

Model	Accuracy	Precision	Recall	F1-Score
Random Forest	74.67%	70.41%	52.86%	33762.05
XGBoost	81.76%	57.59%	59.66%	33762.05
LightGBM	77.72%	73.18%	56.16%	33762.05

Why XGBoost?

We selected XGBoost as our final model due to:

• Highest accuracy (81.76%)
• Better handling of complex feature relationships
• Efficient prediction time
• Good balance of performance metrics

Feature Engineering

Key features used in the model:

features = [
    'MEM_GENDER_ENCODED',
    'MEM_RACE_ENCODED',
    'MEM_ETHNICITY_ENCODED',
    'MEM_AGE_NUMERIC',
    'DIAGNOSTIC_CONDITION_CATEGORY_DESC_ENCODED',
    # Disease flags
    'HAS_HYPERTENSION',
    'HAS_DIABETES',
    'HAS_RENAL_FAILURE',
    # ... and more
]

Model Training Process

1. Data Preparation

   • Feature encoding
   • Handling missing values
   • Data normalization

2. Model Development

   • Cross-validation
   • Hyperparameter tuning
   • Performance evaluation

3. Model Optimization

   • Feature importance analysis
   • Model compression
   • Inference optimization

For detailed implementation and analysis, check our model development notebook.

---

💻 Application Workflow

graph TD
    A[User Input] --> B[Data Processing]
    B --> C[Risk Assessment]
    C --> D[Database Storage]
    D --> E[Dashboard Analytics]
    E --> F[Visualization]
    C --> G[LLM Analysis]
    G --> H[Medical Recommendations]

    subgraph "Backend Processing"
    B
    C
    D
    end

    subgraph "Frontend Display"
    E
    F
    H
    end

Loading

🔒 Security Features

• Secure database connections
• API key management
• Error logging and monitoring
• Data validation and sanitization

📈 Dashboard Analytics

The dashboard provides:

• Risk level distribution trends
• Condition prevalence analysis
• Demographic insights
• Prediction confidence metrics
• Historical data analysis

👥 Author

Ajito Nelson Lucio da Costa

---

Built with ❤️ in Timor-Leste 🇹🇱