Page 265 - Introduction to Statistical Pattern Recognition

5 Parameter Estimation 247

TABLE 5-12
COMPARISON BETWEEN CONVENTIONAL AND
BOOTSTRAP METHODS

..
EL
Data N
Mean Standard Mean Standard
(%)
- deviation (%) (%) deviation (%) (%)
24 17.08 4.89 13.54 3.14 0.1 1
40 13.38 6.04 7.63 3.88 0.07
1-1 80 11.19 2.47 4.06 1.29 0.04
160 11.28 2.35 2.16 1.09 0.03
320 10.67 0.80 0.89 0.37 0.0 1
24 18.33 4.79 14.79 3.86 0.12
40 13.75 3.23 8.88 2.97 0.06
1-41 80 11.19 2.72 4.00 1.56 0.08
160 9.88 1.58 2.28 0.68 0.01
3 20
10.09
- 0.83 0.98 0.40 0.01
24 5 .oo 3.43 4.38 3.02 0.0 1
40 3.63 1.99 1.75 1.21 0.02
I-A 80 2.3 1 1.10 0.88 0.94 0.01
160 2.34 0.90 0.4 1 0.36 0.00
320 2.17 0.48 0.17 0.14 0.00

(a) Conventional L and R error estimates.

Because the samples from each class are bootstrapped independently,
cOv*{yy),yp} = 0.
Using a property of the inverse Fourier transform,

(5.184)

The variance of y)’’ is