Preprocessing tools for nonlinear datasets

Abstract

Two new preprocessing systems developed on the basis of the evolutionary methodology are proposed and used in artificial neural networks (ANN). These systems show a high ability for empowerment of standard ANN model performance when used for prediction/classification problems with complex datasets characterized by nonlinear relations between the variables. Training and testing systems are robust data resampling techniques that are able to arrange the source sample into subsamples that all possess a similar probability density function. In this way, the data is split into two or more subsamples in order to train, test, and validate the ANN models more effectively. The IS system is an evolutionary wrapper system able to reduce the amount of data while conserving the largest amount of information available in the dataset. The performances of such systems were tested in a classification task carried out on two different well-known datasets. The classification accuracy reached by a standard back-propagation ANN model trained first on a random subset and then on subsamples selected by T&T systems, while simultaneously using IS to select the variables, is compared. The results show a significant enhancement of the standard ANN classification ability when the proposed preprocessing systems are applied.