IMPORTANCE OF INFORMATION PREPROCESSING IN THE IMPROVEMENT OF NEURAL-NETWORK RESULTS

This paper compares the success ratio of certain topologies when their input data are changed through different pre-processing methods. It b egins with the database description, and it shows some different kinds of pre-processing that will be applied and the necessary modification s to the input layer of the network. The process is carried out using four networks with supervised learning (Standard Backpropagation, Quic k propagation, Resilient Propagation and Backpropagation with Momentum ) and two with unsupervised learning (ART 1 and Dynamic Learning Vecto r Quantization).