Hi All;
I've used S-plus for Windows v.6 to derive a regression tree that I pruned
to 11 terminal nodes using cross validation. The dependent variable is
forest canopy cover (%), independent variables are a number of terrain
variables like elevation, slope, etc.
From the output I'm given (pasted below) I'm trying to determine how much
(what percent) of the total variance my tree actually explains. How do I
determine this?
Ryan Danby
*** Tree Model ***
Regression tree:
snip.tree(tree = tree(formula = spru2 ~ curv.t + elev + geol + r.sum + rock
+ slp + twi, data = T1, na.action = na.exclude, mincut = 5, minsize = 10,
mindev = 0.01), nodes = c(7., 41., 11., 32., 17., 6., 33., 43.))
Variables actually used in tree construction:
[1] "elev" "r.sum" "rock" "slp" "geol"
Number of terminal nodes: 11
Residual mean deviance: 309.6 = 441500 / 1426
Distribution of residuals:
Min. 1st Qu. Median Mean 3rd Qu. Max.
-71.1300 -9.4300 -0.7222 0.0000 0.6263 70.0100
node), split, n, deviance, yval
* denotes terminal node
1) root 1437 915800.0 15.1500
2) elev<1298.5 615 556300.0 31.0700
4) r.sum<497.5 456 307200.0 24.6900
8) rock<10.5 425 290700.0 26.2600
16) elev<1234.5 338 240400.0 29.0200
32) slp<9.5 208 137500.0 23.6300 *
33) slp>9.5 130 87180.0 37.6500 *
17) elev>1234.5 87 37800.0 15.5500 *
9) rock>10.5 31 990.2 3.0940 *
5) r.sum>497.5 159 177200.0 49.3900
10) rock<0.5 134 138500.0 54.2200
20) elev<1208.5 64 50890.0 65.7500
40) slp<11.5 9 6010.0 32.8800 *
41) slp>11.5 55 33560.0 71.1300 *
21) elev>1208.5 70 71340.0 43.6700
42) geol:2,5,8,9 22 6783.0 18.0900 *
43) geol:1,3,7 48 43560.0 55.4000 *
11) rock>0.5 25 18760.0 23.5100 *
3) elev>1298.5 822 86630.0 3.2280
6) elev<1372.5 189 56490.0 11.6200 *
7) elev>1372.5 633 12860.0 0.7222 *
|