On the Irrelevant Feature Discarding in Neural Networks and its Applications in Medicine
Ion Ciucă, Hord E.D. · Studies in health technology and informatics · 1996
Various applications of neural networks in several areas sometimes involve expensive expertises as for example in medicine etc, while some features seem to have no relevance to outputs. This paper makes some considerations on the possibility of reducing the input space size by discarding the irrelevant features in input vectors based on the input sensitivity. To remove the influence of outputs in computing the respective sensitivities, this paper introduces the normalised input sensitivities and presents a detailed algorithm for discarding the irrelevant features in case of feedforward neural networks.