ENSEMbLe

Inspiration

What it does

The project explores crop yield prediction modelling approaches with validation across on out-of-sample locations and annual observations. It tries to understand how predictors and their importance evolve during crop growth season.

How we built it

By training and stacking SVM, GBM and BrNN models. Then by deriving information on most important predictors and how they change along crop season.

Challenges we ran into

I would explore more appropriate methods for spatio-temporal sampling. Limited time and computing power.

Accomplishments that we're proud of

Transforming variable importance rankings (from permutation-based approach) that provide a better overview of tenmporal dynamics of predictors.

What we learned

Too much coffee may kick..

What's next for ENSEMbLe

Id like to collect feedback from community if the approaches could be improved.

Built With

• qgis
• r