## Inspiration
We were inspired by the nuclear fallout theme where we searched for relevant datasets to this particular theme. We found a Safecast dataset that allows us to forecast nuclear fallout weather.
## What it does
The application gives current nuclear fallout information based on longitude and latitude which is put into the Safecast API to find out how the radiation is in your area. The application can also predict future forecasts using a XGBoost model that has ~99% accuracy.
## How we built it
We built the model using Sklearn and PyTorch (initial and final model) along with the API being in Flask and the front end being vanilla HTML/CSS. For the pre-processing, we used conventional data manipulation libraries like Pandas.

The model itself was initially built as a Random Forest but we eventually went with XGBoost where we increased accuracy from ~60% to ~99% to make it much more reliable.
## Challenges we ran into
We didn't specify requirements in a specific way so people were left confused at times, wasting time. There was also the issue of the random forest model being too inaccurate which was solved by swapping to XGBoost and extra pre-processing. There was also the fact that none of us were particularly good at front-end development and the learning curve for React ended up being too steep (hence why we swapped to raw HTML/CSS).

An obvious challenge was the sleep deprivation from a 24-hour hackathon. We dealt with it reasonably well considering the extra circumstances.

## Accomplishments that we're proud of
We created a highly accurate model that can predict nuclear fallout weather changes based on the Safecast API and everybody in the team had a noticeable role to play.

## What we learned
We learnt more about machine learning and the importance of pre-processing as well as the importance of picking technologies wisely instead of jumping into steep learning curves that wouldn't have worked with a team with different skill levels. We also learnt about how to resolve merge conflicts and some Flask.

## What's next for RandomArborists
We will keep planting random forests (if they work!) and creating cool stuff with machine learning as 3 of us are AI students. The Lorax would be proud.

Built With

Share this project:

Updates