🦠 EpiWatch

AI-Powered Epidemic Intelligence & early-warning platform.

Predicting disease outbreaks before they happen — delivering actionable insights up to 5 days ahead of traditional surveillance systems.

---

🚨 The Problem

Every major epidemic follows a dangerous, predictable pattern:
Cases rise slowly → Nobody notices → Cases explode → Panic begins → Government reacts → It is already too late.

Traditional disease surveillance systems generally detect outbreaks only after the damage is done. By the time hospitals start reporting surges, the infection has already spread uncontainably.

There had to be a better way.

---

💡 Our Solution

EpiWatch is a next-generation, AI-driven epidemic intelligence platform built to monitor, forecast, and visualize global health risks in real time.

• 🔮 Actionable Forecasting: Predicts case counts 7 to 30 days in advance.
• 🗺️ Geographic Intelligence: Automatically detects and isolates high-risk outbreak regions.
• 📡 Transmission Tracking: Estimates the R0 transmission rate dynamically.
• ⚡ Proactive Alerts: Issues early warnings up to 5 days before traditional systems.
• 🤖 Automated Insights: An integrated AI engine dynamically translates raw mathematical model outputs into readable human intelligence.

---

🤖 The AI Engine (3 Integrated Models)

EpiWatch relies on a trifecta of machine learning models to analyze time-series data, classify risk parameters, and evaluate transmission vectors.

1. Case Forecasting Model (lstm_model.pkl*)

• Architecture: Time-Series Forecasting (Facebook Prophet)
• Goal: Predict future epidemiological curves with strict upper and lower confidence bounds.
• Key Regressors: Vaccination coverage rate (vaccination), population movement (mobility_index), and minimum case baseline.
• Accuracy: 94.2% on holdout test data.

2. Hotspot Detection Model (xgb_model.pkl*)

• Architecture: Multi-class Classification (RandomForest Classifier)
• Goal: Categorize global regions into 🟢 Low, 🟡 Medium, or 🔴 High-risk zones based on 10 dense epidemiological features (growth rate, population density, testing ratio, hospital capacity, etc.).
• Risk Calculation: Weight-adjusted formulation prioritizing case growth (40%), mobility (30%), population density (20%), and vaccine deficit (10%).
• Precision: 91.2% (F1 Score: 0.89)

3. Transmission Rate Estimator (gb_model.pkl*)

• Architecture: Regression (RandomForest Regressor)
• Goal: Predict the real-time R0 (basic reproduction number) to dictate whether a disease is spreading (>1.0), stable (=1.0), or declining (<1.0).
• R² Score: 0.86

* Note: Model files are managed externally via Google Drive due to GitHub size constraints. See the Setup Instructions below.

---

📊 Data Pipeline

EpiWatch aggregates and cleans data from trusted global health providers:

Dataset	Source	Purpose
COVID-19 Time Series	Johns Hopkins CSSE	Core training data for Model 1 (Forecasting)
Global Vaccination Data	Our World in Data (OWID)	Vaccine/mortality feature engineering
Mobility Reports	Google	Transit, retail, and workplace movement impact analysis

---

🖥️ Platform Features (React Frontend)

The EpiWatch dashboard is engineered with a "Biohazard / CDC Emergency" dark theme to emphasize urgency and clarity. It consists of 6 core interactive modules:

1. Global Dashboard (/) — Live statistical overview, Prophet trend predictions, risk distribution donuts across 194 regions, and active outbreak tickers.
2. Country Analysis (/country) — Deep-dive analytics featuring Interactive Simulators (drag a slider to see how increasing mobility or vaccination impacts the predictive curves), What-If scenarios, and automatic text-based AI summaries.
3. Global Risk Map (/map) — An immersive Leaflet.js geographic layout highlighting high-risk zones with pulsing markers and real-time bounding box statistics.
4. Hotspot Detection (/hotspots) — Direct output from Model 2 revealing the top 10 most critical regions, supported by 7-day sparkline trends and confidence scores.
5. AI Insights (/insights) — Transparent model diagnostics (SHAP values, accuracy over time) coupled with an NLP Chatbot for querying epidemiological data.
6. Country Comparison (/compare) — Interactive side-by-side radar charts and trajectory comparisons between two selected nations.

🎬 Demo Mode: EpiWatch includes a built-in, fully scripted Auto-Pilot sequence that automatically navigates through all 6 pages for seamless live presentations!

---

🛠️ Technology Stack

Backend Engine

• Framework: Python, FastAPI, Uvicorn
• Machine Learning: Scikit-Learn, Facebook Prophet, Joblib
• Data Engineering: Pandas, NumPy
• Environment: Python-dotenv

Frontend Client

• Core: React 18, Vite
• Styling & Motion: Tailwind CSS, Framer Motion
• Visualizations: Recharts, Leaflet, React-Leaflet
• Networking/Routing: Axios, React Router v6

---

🚀 Setup Instructions

Prerequisites

• Python: 3.10+
• Node.js: 18+
• Git

1. Download the AI Models (Crucial Step)

Due to size limits, trained .pkl models are hosted externally.

1. Download them from this Google Drive Link.
2. Create a models/ directory inside the backend folder.
3. Place lstm_model.pkl, xgb_model.pkl, and gb_model.pkl inside backend/models/.

2. Start the FastAPI Backend

git clone https://github.com/tapshyamangal567/Epiwatch-.git
cd Epiwatch-/backend

# Install Python requirements
pip install -r requirements.txt

# Run the Uvicorn ASGI server
python -m uvicorn main:app --reload

The API will be available at http://127.0.0.1:8000. Swagger documentation is auto-generated at /docs.

3. Start the React Frontend

Open a new terminal tab:

cd Epiwatch-/epiwatch

# Install Node modules
npm install

# Start Vite development server
npm run dev

The dashboard will be available at http://localhost:5173.

---

🔌 Core API Endpoints

Method	Endpoint	Description	Model Engine
GET	/api/predict/cases	7-day future case forecasting globally	Prophet
GET	/api/predict/cases?country=[X]	Specific forecasting by region	Prophet
POST	/api/predict/r0	Real-time R0 transmission rate calculation	RandomForest
GET	/api/hotspots/	Returns the highest risk classified zones	RandomForest
POST	/api/chat	NLP-based chatbot response generation	Rule-based + Data

---

EpiWatch — Predicting tomorrow's outbreaks, today.

Built for the CodeCure AI Hackathon (Track C)