jschulberg
commented
2 years ago The team used a Decision Tree classifier, and achieved an accuracy of 84.7%. When taking a look at the first few layers of the decision tree, we found surprising results. Even though the accuracy scores were generally high, the variables used were surprising. As we can see in the below visualization of the decision tree, the first variable that’s considered in splitting the data is needs_play and the second variable is new_this_week. We generally expected the new_this_week variable, which is just a note taken when a dog is new (but not necessarily that it’s still considered new when adopted), to be unimportant in our analysis.
With these variables of interest, the Decision Tree classifier achieved an accuracy of 84.7% and the following decision tree:
Confusion Matrix completed and tree map completed.