Text classification using the σ-FLNMAP neural network

A neural network, namely the sigma fuzzy lattice neural network with mapping (σ-FLNMAP), is presented and applied to classification of text (documents) from the Brown Corpus benchmark collection of documents. The σ-FLNMAP is presented as an enhanced extension of the fuzzy-ARTMAP neural network in the framework of fuzzy lattices. An individual σ-FLNMAP´s classification accuracy is improved by training an ensemble of σ-FLNMAP modules on different permutations of the training data. Several different vector representations of a document are employed. The results, in a series of experiments, compare favorably with the results by other classification algorithms including K-nearest neighbor and naive Bayes classifiers