Predictive Analytics for Disaster Survival in Educational Gaming

After winning the Most Popular Award in a data communication competition, I collaborated with the host government organization to build a new version of prototype. I designed a web game valued at $17,000 for the Taiwanese market, aimed at disseminating disaster reduction knowledge, where I also conducted statistical validation on a neural network model that identified 256 distinct groups using survey data.
MVP Github Repo Link: github.com/CaslowChien/Disaster-Survival-Rate-Web-Game
MVP Game Link (Click Here to Play)

Due to the confidentiality of the project, the final Github repo is kept private.

Project Results

Developed a $17k data-driven game prototype for disaster risk education, using a deep neural network with PyTorch to classify post-typhoon household survey data and predict survival outcomes with 80% accuracy.
Analyzed and cleaned the data with Pandas, identifying key prevention actions with strong positive correlations at 99% confidence with survival rates in both in linear and non-linear relationships.
Led a cross-functional team to achieve a 70% engagement increase and win “Most Popular Award” in MVP demo day.

Proposal Slides

The final presentation is unavaliable to the public due to the contract protection

UI / UX

I developed the skills to design a website that effectively engages users and enhances their retention.

EDA

This project taught me how to enhance the value of data by developing websites for data presentation, generating predictions through machine learning models, and effectively communicating insights to professionals across various disciplines.

Model & Validation

I applied hands-on statistical validation techniques to real-world data, including methods like cross-validation.