mattprout
commented
6 years ago Adjusted the focus of the model to improving the outcome of all animals by providing a recommendation on how to make the animal more adoptable. The model provides a probability score for the improved outcome after enacting the recommendation. That way those running shelters can use the scores to determine which animals to focus on if resources become scarce. Added an ability to populate attributes about the animal if the animal is microchipped, and the national database happens to contain information about the animal.
This tool is used by animal shelters to predict the likelihood of animals being returned, adopted, or euthanized. These predictions will be used to minimize the numbers of animals that are euthanized, by directing shelter resources to animals most likely to be euthanized.
This is a great idea. The estimate of 1.5 million euthanized animals out of 6.5 million painted a clear picture of the problem. I also like all of the classifiers you laid out. It is easy to imagine how those might all impact the results of the model.
I think this solution could be strengthened if the model predicted which animals should receive shelter resources, rather than which ones are most likely are to be euthanized. I imagine shelters have very limited resources. In a basic example, image a shelter has three dogs that are young, middle aged, and old, and the shelter can only direct its resources to one dog. The best outcome could potentially result from directing resources to the middle aged dog, even though the old dog is the one most likely to be euthanized. I think you start to get at this idea when you discuss sick animals being treated by vets, and behavioral issue animals being given training. These are instances of animals that could potentially benefit from some attention, more than other animals may be able to. Maybe the model will direct the resources to younger animals first, if it finds that certain older animals are less likely to benefit from additional attention? So, maybe rephrase the tool as a model to guide resources, rather than predict the likelihood of being euthanized?
Continuing on the idea of limited resources: It sounds like someone has to enter the information of each animal into this tool. I image each shelter already has a database where someone has entered this information. Maybe this tool could connect to that database, or find some other way which could prevent the employees from duplicating their data entry efforts.
Awesome idea. This sounds like something that could realistically be made, and have a positive impact.