Demystifying Decision Science
Howard Friedman
Adjunct Professor at Columbia University
Simplicity has power
Interpretability matters
Many real-world problems involve binary or categorical predictions:
Logistic regression is highly versatile
Probabilities support decision-making
Easy to interpret and implement
Start with a strong baseline model
Feature transformations can boost performance
Match model assumptions to the data
Use transformations to improve early models before jumping to complex ones
Demystifying Decision Science