1/4
Looks like no tags are added yet.
Name | Mastery | Learn | Test | Matching | Spaced |
|---|
No study sessions yet.
what do classification model output
a score, ranking observations by likelihood of responde (for some models interpreted as probability estimate)
performance depends on threshold
different tresholds → different confusion matrices
evaluating performance conditional on threshold =/= evaluate score ranking
profit-curve : pro/con
pro : interpretable, allows profit optimization
con : sensitive to c&b + class priors
ROC-curve pro/con
pro : independent of class proportions and costs and benefits, AUC
con : interpretability for stakeholders
Cummulative responde or lift curve pro/con
pro : more intuitive than ROC
con : not robust to changes in class proportions