The Polar Bears

Inspiration

We all want to participate in the Kaggle competitions and agreed to the sea ice challenge.

What it does

Predict the velocity of the buoy using a DNN (Deep Neural Network) and XGBoost.

How we built it

We tried linear regression, XGBoost and deep neural network to predict the velocity components. XGBoost comes from the xgboost package and the DNN was built with tensorflow. We formatted the velocity components into velocity magnitude and direction to see how they were working. The models performed better for the original format which is u and v component of velocity.

Challenges we ran into

Both models had trouble with come of the peaks in the sea ice data.

Accomplishments that we're proud of

Working with climate data as it was a new domain for some of us.

What we learned

Climate data peaks are hard to model!

What's next for The Polar Bears

To predict the peaks better we can integrate two models where in the first model we will classify the data into categories and in the second model predict the velocity components based on the classification. It will reduce their tendency to merge to average. And also using some other open source data to make the model better.

Built With

• jupyter
• python
• tensorflow
• xgboost