Please enable JavaScript.
Coggle requires JavaScript to display documents.
Communicating Model Insights (Six types of information communicated during…
Communicating Model Insights
Six types of information communicated during a presentation
areas where the model struggles
sometimes the problem might lie withe the data - might need to argue for more data
most predictive features for model building
directional results make intuitive sense
develop story around feature directionality and why
model quality metrics
discuss cleaning, procuring data, removing data, and why
begin with confusion matrix, annotate for audience
positive predictive value, probability distribution thresholds, extract predicted probabilities, possible proposals
convince management that the model would ultimately save money
feature types especially interesting to management
four types of features to consider before meeting
features requiring further examination
might need further analysis from SME
features which are likely to raise follow-up questions
consider shortening feature names for clarity
immutable features
features good for modeling but of no value to management implementation
management team cannot change/impact these features
features needing to be changed, requiring a re-run
general rule to re-run if features change
mutable features
features that management could potentially change
BUT: management interference with these features can cause complications in model
need to consistently monitor model performance after changes implemented - might need to utilize new data
business problem
should have been refined as autoML process carried forward
also refined when talking to subject matter experts
recommended business actions
should contain explicit recommendations for next steps
Unlocking Holdout
used to check that there havent been mistakes made in the model creation process
best outcome is that the model rankings dont change when holdout is unlocked
small samples more likely to lead to issues