Note! Our final report can be found here: https://drive.google.com/file/d/1FbKZvqoc1nfNUxBMax9jJg3UYBJvFXjN/view?usp=sharing
What we did and why we did it!
Our model lets farmers predict crop yield months in advance of harvesting, reducing logistical stress. We predicted using EVI and temperature data alone.
How we built it, challenges, and accomplishments
We used an ensembled NN model with Random Foresting baseline. We got great meaninful data and highly accurate predictions! High prediction accuracy with minimal usage of CPU space was great! All 4 of our models performed well on both our training and validation set. Through this project, we have all gain valuable insight into the agricultural industry and the statistics behind it, which would be very useful when doing similar projects. Tackling this open-ended question in general has intellectually challenged us, and completing a project is one of our biggest accomplishments.
What we learned, what's next?
Implicit correlations for crop yield are encoded mostly in termperature and soil moisture data. Using a combination of models to predict the outcome, compared using a variety of metrics to avoid biases. Using additional research into the field to interpret our results.
To make it better, using more resources and publicly available data would have been great. NN embeddings for geospatial data and more data analysis on final model.
Also definitely plan on doing more hackathon. This was the first and loved it!
edit: This is just something I found after the fact, our model generally predicts yields within 3% of actual yield.