Growing up in California, I've noticed the devastating impact that wildfires wreak on thousands of families every year. From damaging acres of crops, destroying homes, and displacing locals, wildfires leave many regions in bleak hardship. With changes in climate creating warmer and drier conditions, wildfires are becoming increasingly severe.
After hearing about the disastrous consequences that wildfires wreak on my nearby communities, I was inspired to utilize my skills in computer science to build an app for wildfire victims.
What it does
Safefires provides the latest wildfire updates, a map to learn more about nearby events, predicts wildfire risk based on weather, and displays tips for wildfire prevention and safety. On the news page, users can find recent updates on nearby wildfire containment, how neighborhoods are currently being affected, and any safety alerts. The map includes a visual display of nearby fires, links a page to details, and also directs users towards a donation page for those affected by an ongoing fire. On the dashboard, users can find tips for wildfire safety and a live wildfire risk prediction (low/medium/high and percent risk) based on four factors: temperature, humidity, wind speed, and precipitation.
How I built it
As an iOS app, Safefires was built using Swift and prototyped in Figma. To create the machine learning model for detecting wildfire risk, I trained a support vector machine with UC Irvine's dataset of 500+ wildfires.
Challenges I ran into
Since this was my first time writing code to build a machine learning model (whereas I had used simpler methods with CreateML in the past), it was definitely a challenge to debug my code and improve accuracy. Although I went into this hackathon unfamiliar with machine learning models like support vector machines, I persevered with the help of online resources like Medium.com articles and YouTube tutorials.
Accomplishments that I'm proud of
As someone who starts many projects without pulling through to finish them, I'm proud of working from start to finish and reaching an end result. Furthermore, I'm proud of writing my first machine learning algorithm despite the many difficulties I faced.
What I learned
During this hackathon, I've learned how to use ski-kit learn in Python to create my own support vector machine. More generally, I also learned how to make better use of online resources like reading Towards Data Science articles for brief yet effective machine learning advice.
What's next for Safefires
As Safefire's machine learning model is still a work-in-progress and has the potential to improve on making more accurate predictions, I plan on gathering more wildfire data and experiment with alternate machine learning models for better predictions. Furthermore, I plan on adding a variety of mini-courses for users to learn wildfire prevention and safety from, such as step-by-step walkthroughs for creating an emergency kit and testing drills.