en
Predictive Analytics in Commerce
Jorge Mantilla
02. juil. 2016 19:31 heures
Chapitre 5 › Pruning the Tree:
Evaluating Performance Measurements
Afficher le devoir
Masquer le devoir
Evaluating Performance Measurements
Below you find two models and some corresponding performance measurements. If you should choose one of the two models based on these measurements, which one would you choose? And why?
Model A
CP = 0.0001
R^2 = 0.98
Xerror = 0.65
Model B
CP = 0.01
R^2 = 0.80
Xerror = 0.55
Model B
1
0
I would choose model B, since I'm looking for a tree that has a low number of splits if possible (higher complexity parameter compared to model B) with the ability to predict accurately for different sets of data (relatively small Xerror) and with relatively high predictive power R^2. in other words, Model B does not overfit, the tree stays understandable and has good predictive power.
Perfect explanation Jorge!