Inspiration
There are a lot of prediction-based models that are used to forecast when natural disasters occur, but we still hear about so many tragic deaths and destruction caused every year by hurricanes, tornadoes, and typhoons. We wanted to create an app that not only predicts how a disaster will unfold but also assists people caught in that area so that they can get assistance on finding safety.
What it does
SafePath is a web application that provides real-time alerts and pathing to users who are in close proximity to disasters. It also predicts where the disaster will go and potential damage, and determines new paths to safety in the event of interference. By providing the user with routes and notifications about disasters, we hope that many lives can be saved.
How we built it
We used React as our framework to create the front end of the application since we could have more flexibility with adding animations and making the UI user-friendly. There was a large focus on minimalism and avoiding cluttering up the map, because, in the event of an emergency, the user wants fast and easily digestible information. For the backend portion, we used Flask and Python to create our pathfinding algorithm, and Keras + Python to create our DNN(Deep Neural Network). By training the DNN through API information from openweathermap and the Google Maps API, we were able to predict how a disaster would move on a map, and how a path could be built around it.
Challenges we ran into
A major challenge we ran into was getting good accuracy with our DNN model. There are a lot of factors that goes into determining where and when a disaster will occur and what direction it will unfold in, and we only had two APIs that could give us enough information to train our model off of. However, a lot of the time there are environmental factors and destruction that cannot be accounted for with existing technology, so finding reliable data is something that would require more exploring and possibly hardware based solutions. Another challenge we ran into was creating an algorithm that was efficient enough to pathfind off the predictions and update the user. Since it is a survival-based app, we cannot take up too much time, so it was important we were efficient and minimized overall recalculations of potential safe routes to take.
Accomplishments that we're proud of
We are proud of managed to finish a demoable version of SafePath, which can navigate across a map and show path calculations to safe zones based on a disaster(a tornado) occurring in the area. Working with neural networks and API integration was something new for our team, so it was very fun learning about how it worked and successfully integrating it into our final product.
What we learned
We learned a lot about AI and ML model training, as well as how various frameworks such as tensorflow and keras work, and how their usages can vary based on the type of model and data you want to train/utilize. We also learned a lot about Frontend development and were able to explore a lot of fun and unique libraries that improved the overall user experience.
What's next for SafePath
Going forward, we want to create a more reliable pathfinding algorithm that is more efficient and can provide faster updates to the user. We also want to flesh out our disaster detection system and try to incorporate predictions and pathfinding for multiple types of disasters. Furthermore, the current scope of how far we can predict with our model is something that can be improved upon, as we seek to be able to look over entire states and regions of countries in the future.
Log in or sign up for Devpost to join the conversation.