Page 258 - Introduction to Statistical Pattern Recognition

240 Introduction to Statistical Pattern Recognition

Fig. 5-5 Bootstrap sample generation.

There could be many possible ways to generate artificial samples. One
is to generate samples normally around the existing samples. This leads to a
nonparametric technique called the Parzen approach with the normal kernel
function, which will be discussed extensively in Chapters 6 and 7. In this sec-
tion, we discuss the case hi which samples are generated randomly at the
points where the existing samples in S I are located.

Bootstrap errors: Let us assume that S I consists of N I o1 -samples,
X\'), . . . ,X#!, and N, m2-samples, XI2', . . . ,X$j. Then, we may express the