Neural network model for multidimensional data classification via clustering with data filtering support

Abstract

The paper introduces a neural network model for multidimensional classification via clustering with data filtering support that consists of two neural networks. The first neural network based on Pulse Coupled Neural Network (PCNN) solves dimension reduction and generates appropriate number of features for final classification. The second neural network Projective Adaptive Resonance Theory (PART) solves classification via clustering. The clustering usage is very effective in this case because the proposed model after a small modification of clustering algorithm allows filtering of unwanted data. It means that the proposed neural network model is sensitive to predefined number of classification classes only and all other data that do not belong to the predefined classes are filtered in to separate cluster.