Collection 2004

O ANALIZA COMPARATIVA A PERFORMANTEI ASOCIATORILOR MULTINIVELARI SI A MODELELOR NEUROMIMETICE BAZATE PE FUNCTII RADIALE IN SARCINI DE CLASIFICARE

Written by Petru Lucian CURȘEU, Mircea MICLEA, Mircea Alexandru BIRT on . Posted in Volume VIII, Nr. 2

ABSTRACT

A review of the literature related to neuromimetic modelling of classification mechanisms in areas like medicine, organisational settings and psychopathology reveals their interesting computational performances. On a sample of 80 patients of unipolar, bipolar depression and dysthimic disorder we have compared the classification performance of multilayer-perceptions and radial basis function neural networks. We found that both neural network models outperform regression analysis. An increase in the complexity of the neural network induces an improvement of classification performance. We found also that multilayer perceptions are more suitable for situations when input data are missing and / or the 'noise level' (i.e. measurement error) is high.

KEYWORDS: neural network, classification, psychopathology, multilayer perceptions, RBF neural networks.