Inspiration
Hearing about the damages caused by landslides, the group sought to train an AI model to predict the most vulnerable areas of the West Coast to help spread awareness and inform locals and local authorities of the possible danger.
What it should've done
Trained AI would've classified different geographical regions as high or low risk. This data would be visible in a scrollable map in a web app along with allowing the user to see all of the spatiotemporal data utilized in the making of the classifications.
How we built it
We collected spatiotemporal data from online resources or from deriving from other data, including population density, slope degree, slope aspect, elevation, distance to faults, distance to rivers, land cover, and landslide data. We preprocessed and formatted the data, then used a support vector classifier to process the data and build a classifier model. We made a react web app to visualize the collected information, as well as the model's output, in a leaflet's interactive map.
Challenges we ran into
We ran into two MAJOR challenges. (1) the trained support vector classifier performed horribly; And (2) We were running out of time when we found the method we were importing our data into the web app could not get past an authentication for some of the data. With the trained AI being the core of our project, a large portion of our project's features were incomplete. Also with the time provided, we could not dive into the temporal aspect beyond seasonality.
Accomplishments that we're proud of
We are proud of finding quality data, preprocessing it, formatting it, and putting it in the web app in the time we had. Also, the size of the dataset was impressive. We processed data hundreds of thousands data points.
What we learned
We learned that a support vector classifier might not be the ideal model for this dataset. We also learned that it is not realistic to complete a research project of this scale in 24 hours.
What's next for Landslides Susceptibility Classifier
The data is there. After identifying a new AI model, fixing possible issues with the data, and a good amount of testing and tinkering, this could become a tool that can orient residents and governments to avoid risks associated with landslides.
Log in or sign up for Devpost to join the conversation.