Practical Neural Network Recipies in C++

412 Chapter 22

Confusion Matrices

After a neural network (LAYER or KOHONEN) has been trained as

a classifier (in CLASSIFY mode), its performance can be evaluated

with the aid of a confusion matrix (page 348). This matrix is evaluat-

ed one row at a time. A row consists of as many elements as there are

output neurons, plus one for the reject category.

The first step is to zero the counters in the row. This is done

with the command RESET CONFUSION:. We must also set the

classification threshold. This command is CONFUSION THRESH-

OLDtnumber, where the number is a percent (0-100) of full activation.

Once set, it will remain at that value unless