A new approach of classification for non-Gaussian distribution upon competitive training

In this paper, we present a new neural and statistical classification approach. This procedure uses the neural network with competitive training to detect the local maxima of the probabilities density function's (pdf) which are considered as the prototype of the classes in the data distribution. In order to take account of different forms of distributions; Gaussian and non-Gaussian, we used as criteria of resemblance the Mahalanobis distance that takes into account the dispersion of the distributions.