A crop recommendation tool for the novice organic farmer.
The weather data was aggregated from over six thousand NOAA weather stations throughout the nation.
The farms data was aggregated through listings at the Department of Agriculture -- this ensured that they were officially certified as organic farms, under the set standards and requirements.
The model was built using a one vs all classifier for each target label; several types of classifiers were tested: decision trees, random forest, SVC.
The model produces a prediction probability for each target label (i.e. crop / product) since essentially, each target has its own classifier. Then, the probability predictions are sorted to get the ten largest probabilities, which correspond to the rankings you see returned.
The web application is built using the Flask framework deployed with Heroku. Details at: Stackshare.
The project was a three month long effort, which encapsulated various stages of the product development pipeline: industry research, data collection, data preprocessing, model experimentation (machine learning), product design, web application development, & user testing.