Análisis de datos con redes neuronales aplicadas al diagnóstico de la solvencia empresarial: (Sector Societario Ecuatoriano)

Abstract

The purpose of this work is to present neural networks as an alternative tool for the diagnosis of the solveney situation of Ecuadorian firms. The phenomenon of the firms' solvency is analyzed quantitatively and qualitatively, classifying to the companies in solvents or insolvents. In the quantitative analysis, three types of predictive neural network models are compared: multilayer perceptron, radial base function network and Bayesian network; and two statistical methods,  discriminant analysis and logit analysis. For the qualitative analysis, the self-organizing maps of Kohonen or Kohonen neural network is employed to classtty patterns; this procedure is evaluated in comparison with a well-known statistical technique known as multidimensional scaling.