🌟 Inspiration
Every monsoon season, India faces a surge in fever-related illnesses like dengue, chikungunya, and enteric fever, overwhelming hospitals and diagnostic labs.
We were inspired by a simple yet powerful question — can AI help predict outbreaks before they happen?
That idea led to OutbreakCast-Fever Forecasts, an AI-driven solution that transforms health and weather data into early warning signals for better healthcare preparedness.
💡 What it does
OutbreakCast uses machine learning and time-series forecasting to predict fever outbreaks across Indian districts up to 4 weeks in advance.
It analyses rainfall, temperature, and historical case data to identify future risk patterns.
Each district is classified into a risk band:
- 🟢 Green: Low risk
- 🟠 Amber: Moderate risk
- 🔴 Red: High risk
An interactive Stream-lit dashboard visualises these predictions, helping health officials and hospitals respond proactively.
🏗️ How we built it
We collected open datasets from IDSP, NCVBDC, IMD, and NASA IMERG, combined them with population baselines, and cleaned them into a district-week format.
Using Python (Pandas, NumPy, scikit-learn, XGBoost), we engineered lag features, seasonal patterns, and rolling averages for accurate time-series modeling.
Our forecasting equation is based on lagged relationships: [ \hat{y}{t+h} = f(y{t}, y_{t-1}, X_{t}, X_{t-1}) ] Where:
- ( y_t ) = current fever cases
- ( X_t ) = weather features (rainfall, temperature)
- ( f(\cdot) ) = trained XGBoost regression model
The final results were visualised through a Stream lit dashboard that loads forecast CSVs and generates live insights.
⚙️ Challenges we ran into
- Finding clean, district-level outbreak data from government portals
- Managing missing or delayed data reports in real-world surveillance systems
- Designing lag features and handling seasonal variations in small datasets
- Balancing model accuracy with interpretability for healthcare users
🏆 Accomplishments that we're proud of
- Built a fully functional outbreak forecasting pipeline within hackathon time limits
- Achieved realistic 4-week-ahead predictions using only public data
- Designed an interactive, user-friendly dashboard
- Successfully demonstrated the potential of AI for preventive healthcare in India
🧠 What we learned
We learned how AI can bridge healthcare and data science, how to structure spatiotemporal datasets, and how visualisation improves understanding.
We also realised the importance of data ethics, privacy, and the challenge of modelling public-health data with limited granularity.
🚀 What's next for OutbreakCast-Fever Forecasts We plan to:
- Integrate real-time weather APIs for live updates
- Deploy the dashboard to the cloud (Stream lit Cloud / AWS)
- Expand to more diseases and finer geographic levels
- Collaborate with healthcare organisations for pilot testing and validation
OutbreakCast’s goal is to make disease forecasting accessible, actionable, and impactful for every healthcare system.
Built With
- github
- google-colab
- matplotlib
- numpy
- pandas
- python
- scikit-learn
- streamlit
- xgboost
