Performance analysis of a statistical and an evolutionary neural network based classifier for the prediction of industrial bankruptcy

Corporate failure has been causing not only considerable concern to the banks, government and financial institutions, but also financial embarrassment to the industrial units by seriously threatening their very viability. Therefore developing a model to predict potential industrial sickness as an early warning screen is the need of the hour. Various prediction models using financial ratios as predictor variables and employing statistical techniques or neural networks as classifiers have been proposed. However, these models consider the selection of only a few financial ratios according to a choice based criteria. In this aspect, an earlier investigation by the authors on the application of principal component analysis (PCA) which allows any number of financial ratios as input thereby dispensing with the need to choose selective financial ratios, has turned out to be beneficial. Its ultimate function is to reduce the dimensionality of data sets and represent the same by a set of principal components, which intrinsically represent the information content of all the financial ratios put together. In this study we analyze the performance of two prediction models making use of the principal components of financial ratios as their input but with multiple discriminant analysis and an evolutionary neural network as their classifiers, The performance has been analyzed over financial data of 72 Indian manufacturing companies for the financial years 1998-2001, with 21 financial ratios as predictors