Practical Neural Network Recipies in C++

Confidence

Measures

367

can happen if some nonaircraft objects are present in the sample, but

under-represented in number. With that dire warning firmly en-

trenched, let us proceed.

Computing Confidence

In Chapter

12,

Probabilistic Neural Networks, we saw how probability

distributions could be estimated from samples by means of Parzen

windows. We will use that technique later in this chapter. For now,

though, a much more straightforward method is preferable in that it

does not require arbitrary choice of a smoothing parameter or a

particular window function. Also, note that for computing confidence

measures as described above, we do not reall ...