A formal selection and pruning algorithm for feedforward artificial neuralnetwork optimization

A formal selection and pruning technique based on the concept of local rela tive sensitivity index is proposed for feedforward artificial neural networ ks. The mechanism of backpropagation training algorithm: is revisited and t he theoretical foundation of the improved selection and pruning technique i s presented. This technique Is based on parallel pruning of weights which a re relatively redundant in a subgroup of a feedforward neural network. Comp arative studies with a similar technique proposed in the literature show th at the improved technique provides better pruning results in terms of reduc tion of model residues, improvement, of generalization capability and reduc tion of network complexity; The effectiveness of the improved technique is demonstrated in developing neural network (NN) models of a number of nonlin ear systems including three bit parity problem, Van der Pol equation, a che mical processes and two: nonlinear discrete-time systems using the backprop agation training algorithm with adaptive learning late.