Inspiration
We knew that we wanted to do something with predicting natural disasters, and it was only after ages of searching on various database hosting platforms for relevant information did we find a data set on Kaggle that was perfect for it.
What it does
Simply put, latitude and longitude, it will predict the next earthquake for that region; this includes magnitude, depth, date, and time. The closer you are to a fault line, the more likely it is that there will be an earthquake.
Challenges we ran into
At first, when we trained our model, it was far better at detecting where earthquakes weren't going to be, rather than where they were going to be. We jokingly coined the term "Artificial Stupidity" to describe the 0% accuracy that we ended up creating.
The Keras API was simple at its core, but it was difficult to apply to our specific use-case. We tested out using a recurrent neural network, but it wasn't ideal for the sporadic timestamps that we had.
Accomplishments that we're proud of
Actually getting the model to be accurate was a huge accomplishment for us. Sleep-deprived at 6:34 AM, we saw the model be 98.87% accurate after hours upon hours of trial and error. The fact that this project succeeded is an accomplishment in of itself. We're very proud of the work we did.
What's next for Forequake
As it stands, it's not exactly user friendly. It's friendly for individuals who are used to a CLI or programming in general, but individuals who aren't would be completely lost as to how to use it. Developing a proper GUI would be a good next step for us if we were to continue pursuing this project.
Log in or sign up for Devpost to join the conversation.