ACHIEVING SUPERIOR GENERALIZATION WITH A HIGH-ORDER NEURAL-NETWORK

The HONEST neural network is a recently-developed generalisation of th e well-known Sigma-Pi high order neural network. HONEST has previously been applied to diabetes forecasting and feature combination in art O thello evaluation function. In this paper we apply HONEST to the class ification into age-groups of abalone shellfish, a difficult benchmark to which previous researchers have applied cascade correlation, standa rd backpropagation with a Multi-Layer Perceptron (MLP) network, Quinla n's C4.5, and the DYSTAL network. While the best reported test set per formance by previous researchers is 65.61% correct classification, HON EST was able to achieve 72.89% correct rest set classification. In add ition, HONEST's transparent structure allows us to manually examine th e network state and make observations about the solution the network h as learned.