Summer 2023
Falling Fruit is a Colorado-based nonprofit that celebrates foraging and provides world’s most comprehensive open geographic database of food-producing plants in urban areas that benefit individual foragers, and officials in education, research, urban planning, and art.
With the vast range of edible plants displayed on Falling Fruit’s website, it is hard to determine which ones have harvest ready to be foraged. Our team utilizes Machine Learning to provide time-series predictions that can be used by Falling Fruit to filter out plants that are ripe and utilize the website more efficiently.
https://github.com/falling-fruit/dfg-seasons/tree/master
In the first few meetings with our client, we explored Falling Fruit’s mission and the goals it wanted to achieve with our team at Develop For Good. After looking at existing features and the user analytics, we pursued the following as important considerations and steps for our project:
Phenology Research & Accuracy: We want to base our predictions based on mathematical models that will result in high accuracy results
Taylor et. al (pyPhenology paper) is research we decide to base our models from.
Selecting Data: we decide to collect data for overrepresented species of plants in the Falling Fruit database and predict ripeness in countries that generate heavy traffic to make this feature available to a majority of Falling Fruit users upon implementation
Data Collection: we must collect sufficient data for the models parameters
Adaptation: Models must be adapted to available data from the researched models.
Training and testing models: we must make sure the returned day of ripeness prediction makes sense statistically to reality and user observations
At the end, our model development process is based off this: