Next: A Five Class Problem
Up: Pattern Recognition
Previous: kNearest Neighbor Classifiers
Confidence Intervals
When classifying images using test data, true classifier performance
(P_{c}) is estimated as

(1.4) 
Each classified sample is therefore a Bernoulli trial and the estimate
of classifier performance (
)
is distributed as a binomial
random variable with a mean of P_{c} and a variance of
.
As the number of samples is increased, the
binomial distribution can be approximated by a Gaussian distribution
with the same mean and covariance. It is then possible to assign a
confidence interval to the performance estimate,
[18, p. 250]

(1.5) 
where z_{u} satisfies

(1.6) 
for a particular value of u (u=0.95 and z_{u}=1.96 below, unless
otherwise stated).
Copyright ©1999 Michael V. Boland
19990918