Fire Predictor

Inspiration

Due to climate change, wildfires have been getting much worse as evidenced by the recent fires in LA. Therefore, there are bound to be more so the best thing that we can do is come up with plans to contain them as soon as they are discovered so they do not spiral out of control.

What it does

Our application takes user input in the form of lat/long, month, cause, and the year of a fire that has been discovered and returns a detailed plan of what should be done to ensure it can be contained

How we built it

We pre-processed our data from the kaggle dataset using one-hot encoding and scaling techniques. From there, we ran a neural network on the model and saved it as an h5 file. From there, we built out a frontend that accepted the same inputs as the model did and used that to get model output. The model output was then passed into a groq api which called llama 3 model to come up with the best way to contain the fire based on the size of it we predicted. From there, we displayed that in a responsive UI to the user.

Challenges we ran into

Some challenge we ran into were training the dataset since it was so large, so each epoch took a really long time. Additionally, integrating the model into our full stack app took a while because we had to convert the user input and pre-process it exactly like we did in the model.

Accomplishments that we're proud of

I'm very proud that we were able to input model output into the groq api because it made more niche use of it that served the purpose of our project. I'm also proud that for our application, we were able to create an interactive map that let the user input the exact location of the fire discovered and input it to our model which would use that and other inputs to predict the fire size

What we learned

I would say we learned how to store the outputs of machine learning models into different file formats that can be used later to make new predictions on unseen data. We also learned about different neural network architectures and activation functions that can be used to improve training accuracy