Assessments of neural network classifier output codings using variability of Hamming distance

Abstract

Influences of different output codings on the performance of a Back-Propagation Neural Network classifier were investigated. Experimental results obtained show that output codings with equal Hamming distance outperformed output codings with un-equal Hamming distance between the output states. Performances of the network classifiers are also shown to be closely associated to the variability of Hamming distance between states of the codings used.