Neural network approach to improving fault location in local telephone networks

This paper reports an investigation into the possibilities offered by neural networks (NN), in particular, a multilayer perceptron and radial basis function network, for improving fault location in local telephone networks. It shows that NN can model complex relationships between faults and line parameters measured more effectively, and a substantial increase in classification rate has been achieved on a large number of real fault cases. The paper also describes effective data pre-processing methods, including a novel technique for representing symbolic data. In addition the work has demonstrated methodology for using NN in the classification of line test data.