A regression tree on salary-like data: log salary by years of experience and hits (260 players, simulated)

the regions (colour = predicted log salary)

the tree

Grow a large tree, then prune it back: cost-complexity pruning (200 points of a smooth 2-D function)

the subtree's regions

error against the number of leaves

training MSE10-fold CV MSEtest MSE

How good is a split? Misclassification error, Gini index and entropy for one threshold
misclassification errorGini indexentropy / 2class Aclass B

Many trees: bagging, random forests and boosting on Friedman's benchmark (10 inputs, n = 300)