Design of supervised classifiers using Boolean Neural Networks

In this paper we present two supervised pattern classifiers designed using Boolean Neural Networks (BNN). They are (a) Nearest-to-an-Exemplar (NTE) and (b) Boolean K-Nearest Neighbor (BKNN) classifier. The classifiers use the idea of Radius of Attraction (ROA) to achieve their goal. Patterns are classified by constructing hyper spheres in feature space with the exemplar nodes as the centers for the NTE classifier and with nodes representing each of the training patterns as centers in the BKNN. The radii of these hyper spheres are to be adapted to the problem at hand. Both these classifiers are tested with well-known data sets of binary as well as continuous feature values. Results obtained are comparable with those obtained by similar existing classifiers.