STAR SHIELD

Training Visualization
Heatmap highlighting problematic objects
General all-purpose dashboard

Inspiration

We wanted to come up with a direct use of scientific data to a machine learning model that was both relevant to some problem regarding the space theme, and build out software that looked like it was straight out of a childhood superhero show. It gave us the idea to use machine learning on near earth object approach data to classify hazardous and non-hazardous approaches that could then be relevant to long term planetary safety from calamities.

What it does

STAR SHIELD uses data from NASA JPL Close Approach Data (CAD) to approximate whether a space object would come dangerously close to the Earth. It has been trained off of the past 5 years of close approach data to create a Random Forest model that takes the next 3 months of asteroid data (specifically the diameter and velocity among other parameters) to update a stylish dashboard.

How we built it

The backend was made using FASTAPI to build a prediction layer with a python binary classifier model, and a random forest risk prediction model. This layer functioned on top of the NASA REST API provided for JPL CAD to provide predictions about whether a close approach is hazardous or not. It then provides some general metrics about probability of risky approaches based on diameter, distance and so on.

Challenges we ran into

We had challenges communicating what each team member required from the other- mostly specific things, such as what values needed to be passed between the 3 layers- API Query layer, Model layer, and Front-end layer. We eventually made a clear chart of what needs to go where and forged ahead from there.

Accomplishments that we're proud of

We're especially proud of being undergrads who competed in the grad-track. In addtion to this we are proud of how the dashboard came out- it really captures the vibe of what we’re going for and really brings the storytelling aspect together for data visualization of what we are showing our audience. At the end of the day, the purpose of data manipulation is to tell a story- and the dashboard really elevates that aspect of the project.

What we learned

We learned how to properly communicate complex needs between the team in a clear and efficient manner. We also learned how to work on beautiful UI in a manner that we hadn’t really explored before: we never considered the importance of a good-looking application to the user experience much before in hackathons. The hackathon was a detailed study of the amount of data manipulation needed to actually build a model, where we had to multiple levels of manipulation.

What's next for STAR SHIELD

We will try to project where the asteroids land on the Earth and further classify the danger level of these asteroids. For example, if it lands in a densely populated area, it follows that the danger level would be classified as high. Likewise, if the object lands in the ocean, we need to calculate if the impact would be big enough to cause a tsunami/wave large enough to cause human danger. It’s a steep challenge, but with more than 24 hours and our talented team, the stars are the limit ;)