The Systematic Trajectory Search Algorithm for Feedforward Neural Network Training

Abstract

In this work, the systematic trajectory search algorithm (STSA) is proposed to train the connection weights of the feedforward neural networks. The STSA utilizes the orthogonal array (OA) to uniformly generate the initial population in order to globally explore the solution space, then it applies a novel trajectory search method that can exploit the promising area thoroughly. The performance of the proposed STSA is evaluated by applying it to train a class of feedforward neural networks to solve the n-bit parity problem and the classification problem on two medical datasets from the UCI machine learning repository. By comparing with the previous studies, the experimental results revealed that the neural networks trained by the STSA have very good classification ability.