A Meta-cognitive Fully Complex-valued Fast Learning Classifier for real-valued classification problems

Abstract

In this article, a Meta-cognitive Fully Complex-valued Fast Learning Classifier (Mc-FCFLC) for solving real-valued classification problems is presented. Mc-FCFLC consist of two components namely, a cognitive component and a meta-cognitive one. The cognitive component of Mc-FCFLC is a single hidden layer network (FCFLC) with a nonlinear input and hidden layer, and a linear output layer. The meta-cognitive component of Mc-FCFLC consist of a self-regulatory learning mechanism that chooses a best learning strategy among what-to-learn, when-to-learn and how-to-learn for a given sample. The sample is either deleted, used for adding a new neuron or else it is reserved for future use. Thus the architecture of Mc-FCFLC is constructed during the training process. The performance of the Mc-FCFLC is evaluated with the other complex-valued and a few best performing real-valued classifiers on a set of benchmark classification problems obtained from the UCI machine learning repository. Further, a practical acoustic emission signal classification problem has been addressed. Performance results demonstrate that Mc-FCFLC has better classification ability than the other classifiers existing in the literature.