Page 329 - Introduction to Statistical Pattern Recognition

7 Nonparametric Classification and Error Estimation 31 I

TABLE 7-1
3NN ERROR ESTIMATION PROCEDURES

1st NN 2nd NN 3rd NN
- - or

- - Error
0
0
1 2 1 Correct
x3 x IO 23
1 1 I Error
x 1x 25 36

2
2
x35 - x366 - 2 Error
x536
# of errors
A
EL =
N
(a) 3NN (or 3NN leave-one-out)

# orerrors
&R =
N
(b) 2NN (or 3NN resubstitution)

procedure to Table 7-l(b), an error occurs only if the classes of the 2nd-NN
and 3rd-NN agree and differ from the class of the 1st-NN (see XN), because the
class of the 1st-NN is the same as the true class. Therefore, the number of