Structural and smoothing parameter optimization of probabilistic neural network through evolutionary computation and its usage in odor recognition system

Probabilistic Neural Network has received considerable attention nowadays and obtained many successful application. This type of neural system has shown marvelous higher recognition capability compare with that of Back-Propagation neural system. However, this neural system shown some drawbacks, especially on determining the value of its smoothing parameter and huge neural structure when large numbers of data are necessary. Supervised-structure determination of PNN is an algorithm to solve these problems by selecting a set of valuable neurons using Orthogonal Algorithm and determining the optimal smoothing parameter value using Genetic Algorithm. In this paper an experimental set up for comparison of the Supervised-structure determination of PNN with that of the Standard PNN as a neural classifier in the Odor Recognition System is conducted. Experimental results show that the Supervised-structure determination of PNN performed higher recognition rate compare with that of Standard PNN, even using lower number of neurons.