Efficient pattern classification model with neuro-fuzzy networks

The objective of this article is to develop an efficient model of neuro-fuzzy (NF) pattern classification with rough set theory that utilizes the best possible extracted features. Generalization capability of the model is enhanced with a fast and improved learning algorithm called extreme learning machine (ELM). The model explores the features of an input pattern to the possible extent in order to acquire improved class discriminatory information for data sets with ill-defined and overlapping class boundaries. This is obtained through feature-wise degree of belonging of patterns to different classes. Two ways of feature exploration; one using fuzzy set theory to deal with the impreciseness, and other using neighborhood rough set (NRS) theory to deal with the uncertainty and vagueness in the data sets, makes the classification model more efficient. A fuzzification matrix is finally developed with these features, whose elements are fed as input to the neural network. The resultant model thus takes the advantages of fuzzy sets, NRS and ELM algorithm mutually, which is highly suitable for the classification of data sets with overlapping class boundaries. Superiority of the proposed model to other similar methods is established qualitatively and quantitatively using both completely labeled data sets and partially labeled remote sensing images.