Please enable JavaScript.
Coggle requires JavaScript to display documents.
Larsen Ch. 20: Communicating Model Insights (Six types of info for…
Larsen Ch. 20:
Communicating Model Insights
Useful information for understanding the model vs.
useful for making business decisions
Six types of info for presentation
Areas where model struggles
Model quality metrics (confusion matrix)
Business Problem
Most predictive features for model building
Insights for mgmt on feature types
Recommended business actions
Key Things to Note
Determine generalizable relationship b/w features and target
Unlocking Holdout
Observe ranking differences b/w cross-validation and holdout- best model is when ranking does not change when sorting between the two, or remains in top of leaderboard
Reevaluate assumptions made with SME if this is not the case
Focus on business problem
Pre-processing and model quality metrics
Confusion matrix
Holdout sample
Process of procuing data, cleaning data, and modifying to effectively address issues
Algorithms- logistic regressions and deep-learning neural networks
Areas where model struggles
Two types of data: Internal and External
Most predictive features
Four kinds of features to consider:
Features that need more examination
Immutable features
Features that need to be changed and re-run
Mutable features
Recommended business actions
How to implement and pilot programs