Please enable JavaScript.
Coggle requires JavaScript to display documents.
Chapter 20: Communicate Model Insights (Features (show the most predictive…
Chapter 20: Communicate Model Insights
Presentations geared toward the correct audience/ catering to them
1.) Business Problem
revisit and address this is
2.) Model Quality Metrics
3.) Areas where a model struggles
4.) Most predictive features for model building
5.) Feature types especially interesting to management
Unlocking the Holdout
proves that mistakes weren't made
have to resort leaderboard after doing so
the best outcome is one that the order of models did not change between the sorts
6.) Recommend Business Actions
Pre-processing and model quality metrics
begin with the confusion matrix
explains positive predictive value, negative predictive value,true positive rate, true negative rate
provides validity and credibility to the model
could ultimately result in cost savings
Show where model struggles
address that the dataset is not necessarily a predictive dataset
readdress internal and external data
Features
show the most predictive ones
address that not all features are created equally
What needs to be changed
features that require further examination
immutable/ mutable features