Figure 6.

Probabilistic neural network. The left hand side (A) shows the network architecture. Each feature is represented by an input node. Each sample in the training set is represented by a node in the hidden layer. The nodes of the hidden layer evaluate the density value of a new, yet to classify data sample. The nodes in the summation layer sum up the density values for each class. Finally, the output layer outputs the class with the highest estimated membership probability. On the right side (B), a one-dimensional example data set is shown. The blue and the red dots represent training data from two different classes. A chosen probability distribution, in our example a Gaussian distribution, is centered at each of the data points of the training set. The green dot (x = 1.4) is a new data point we want to classify. In our example, the density values of the Gaussians from the blue class are small at the location of the green dot, whereas the density values of the red class are higher, indicating that the green dot is more likely to belong to this class.