Data_Doomdays_Angelina

Inspiration

I’ve always been passionate about global justice, data for good, and understanding how climate change and poor infrastructure collide to create crisis hotspots. With the rise of natural disasters, I wondered: what if we could turn risk scores and historical data into something useful, even predictive?

That curiosity led me to this project—merging INFORM Risk Index with the EM-DAT disaster database to not only validate and visualize historical patterns, but also to forecast future disasters, analyze policy readiness, and recommend smarter resource allocation.

What it does

Data_Doomdays_Angelina is a full-stack, end-to-end data science project that:

• Identifies top high-risk countries based on INFORM scores
• Correlates those scores with real disaster frequency and population impact from EM-DAT
• Predicts disaster likelihood for 2025 using moving averages and regression
• Highlights under-supported countries in terms of aid vs. impact
• Uses clustering and heatmaps to visualize regional risk patterns
• Applies feature importance techniques to reveal which risk indicators matter most
• Offers policy insight on governance, infrastructure, and underreporting

• Builds creative storytelling tools (like a retro “mission briefing”) to engage users

  How we built it

  Solo-built using:

• Python for data processing, regression modeling, clustering, and feature selection

• Pandas, NumPy, Scikit-learn, Seaborn, Matplotlib for analysis and visualization

• Jupyter Notebook for EDA and reporting

• INFORM Risk Index and EM-DAT datasets for ground truth comparison

• Markdown + Storytelling for accessible reporting

• Retro-style visual themes for user engagement (in progress)

Challenges we ran into

• Merging messy datasets: INFORM and EM-DAT had different country names, time frames, and missing values that required intensive cleaning.
• Underreporting issues: Some countries had high risk but no disaster records, making it hard to draw accurate correlations.
• Balancing complexity vs. clarity: Sophisticated models performed better, but were harder to interpret for non-technical users.
• Being a one-person team: Every step—data cleaning, analysis, design, storytelling—had to be done solo, which was both empowering and overwhelming.

Accomplishments that we're proud of

• Successfully built a pipeline that translates global risk indicators into meaningful forecasts
• Identified 5 countries with high risk but low historical data, raising important questions about underreporting or resilience
• Built a visually engaging, story-driven interface to make data feel human
• Created a fully independent end-to-end project from idea to prediction to storytelling ## What we learned
• Not all risk indicators are created equal—some truly drive outcomes, others don’t
• Combining datasets amplifies insight, but also multiplies complexity
• Climate-related risks are increasing sharply, especially in regions like Sub-Saharan Africa and South Asia
• Solo-building is tough, but incredibly rewarding when the vision comes together ## What's next for Data_Doomdays_Angelina Finish building the retro game-style visualization tool that explains disaster risk like a mission scenario
• Deploy the tool publicly so anyone—from policymakers to students—can explore global disaster risks interactively
• Add a real-time update feature that pulls fresh disaster data every quarter
• Write a blog post or mini case study to showcase the work to nonprofits or disaster response organizations
• Explore integrating with platforms like ReliefWeb, UNDRR, or Mapbox for expanded functionality

Built With

• google-colab
• markdown
• matplotlib
• numpy
• open-source-datasets
• pandas
• python
• scikit-learn
• seaborn