Data Doomsday

Inspiration

With climate change intensifying natural disasters and global crises becoming more frequent, we were inspired to answer a critical question:
"Are we sending help where it's truly needed most?"

We wanted to bridge the gap between forecasted risk and historical disaster response, and bring equity and strategy into global aid allocation.

What it does

Data Doomsday is a global disaster analysis and forecasting platform that:

Identifies countries with high disaster risk using INFORM indicators
Analyzes historical disaster impact from EM-DAT (2000–2024)
Detects mismatches between risk and aid distribution
Uses machine learning to predict future disaster frequency and types
Recommends optimized resource allocation based on need, not just visibility
Visualizes clusters, hotspots, and policy gaps across countries

How we built it

Cleaned and merged INFORM Risk Index and EM-DAT Disaster Database
Performed EDA, correlation analysis, and clustering with Pandas and Seaborn
Applied regression models, feature selection, and KMeans for trends and forecasts
Designed storytelling cutscenes and retro game-style interfaces for bonus creativity tasks
All implemented in a single Jupyter Notebook (Colab) for reproducibility

Challenges we ran into

Matching and aligning inconsistent country names and disaster labels
Sparse or zero data in EM-DAT for some high-risk countries
AUC evaluation failed due to multiclass imbalance, requiring metric adjustment
Interpreting geopolitical and policy context behind data imbalances
Creating insights that are both data-valid and policy-relevant

Accomplishments that we're proud of

Built an end-to-end pipeline from data ingestion to prediction and policy recommendation
Highlighted countries most at risk but least supported — a key humanitarian gap
Successfully linked quantitative analysis with qualitative storytelling
Completed all hackathon tasks, including bonus challenges, with clarity and originality

What we learned

Combining datasets from different domains (risk forecast vs. historical impact) requires careful alignment
Data science can expose hidden inequities in aid and disaster management
Visual storytelling can make complex global issues more understandable and actionable
Policy-aware machine learning is both powerful and essential for real-world impact

What's next for Data Doomsday

Build an interactive dashboard with real-time INFORM updates and EM-DAT overlays
Add simulation features for “what-if” disaster preparedness scenarios
Collaborate with humanitarian orgs to validate and refine resource optimization logic
Turn this into a scalable tool for policy analysts, NGOs, and global aid platforms