AquaImpact

Inspiration

As climate change accelerates and freshwater demand rises, countries around the world are under growing water stress. I wanted to build a tool that could predict future water stress using real-world data to help decision-makers see how environmental actions affect water sustainability. The UN SDGs 3, 6, and 13 served as a baseline: clean water access, health, and climate resilience.

What it does

AquaImpact predicts a country's Water Stress Index (0–100) using a machine learning model trained on environmental and emissions data. Users can:

1. Simulate future conditions with sliders
2. Get real-time predictions and a category label (Low, Medium, High)

How I built it

1. Data: I used multiple datasets from the UN Statistics Division and UNFCCC, including freshwater abstraction, renewable water resources, CO₂ emissions, and more.
2. Engineering: I derived 7 core features and manually weighted them to create a stress score.
3. Model: I trained an XGBoost regression model to predict the Water Stress Index on a 0–100 scale.
4. UI: The app was built using Streamlit, with custom sliders and clean UX focused on usability.

Challenges I ran into

The two biggest challenges I ran into were: cleaning and merging data across datasets from different agencies, and choosing weights that fairly represent all stress factors without skew.

Accomplishments that I'm proud of

1. Built a complete, working app with real-world UN data
2. Developed a composite index with transparent logic
3. Created an accessible UI that makes complex environmental data intuitive
4. Tied the project directly to three UN SDGs with relevance

What I learned

1. How to create domain-specific features from messy global datasets
2. Trade-offs between model complexity and interpretability
3. Real-world applications of environmental data science and ML

What's next for AquaImpact

1. Partner with local water agencies or NGOs to integrate AquaImpact into decision-making tool
2. Add real-time data input or CSV upload for researchers and agencies
3. Enable country comparisons and historical stress trends

Built With

• machine-learning
• python
• streamlit
• un-data
• xgboost