Analyze eBird bird sighting data to track migration patterns and build model to determine rareness of given bird sightings.
To read the blog post I wrote on this part of the project and some visualizations I made for it, go here.
If you want to dive into the code, check out the migration_analysis/ folder. I've condensed everything down to 1 Jupyter Notebook.
For the predictive model, I labelled bird sightings in my training set as either ['Common', 'Uncommon', or 'Rare'] based on the percentage of all bird sightings in their region during that season they represented. Then I trained a Random Forest Classifier to predict off new bird sightings.
You can check out the work done for this stage in the rareness_predictor/ folder.
Turned out, a predictive model was the wrong was to go. If someone would enter a bird that was never seen in that region during that season before, I know what I want that to return - Rare!
So I moved to a lookup table model. Here, I loaded the database with the percent of total sightings (in a region during a season) each bird was and return a label based on that.
The table used for this can be found at the end of the prepare_and_label.py module in rareness_predictor/sample/. The database itself is deployed as the backend below.
The backend API serves up rareness labels given a bird, region, and season. Additionally, it serves a list of birds, states, counties, and seasons for input forms on the front-end.
The code for the API is all inside the backend/ folder, although the actual documentation is located here.