Improving the generalisation of the N -tupleclassifier using the effective VC dimension
N. P. Bradshaw, Igor Aleksander · Electronics Letters · 1996
The VC dimension of an n-tuple classifier predicts poorer generalisation performance than is found in practice. The effective VC dimension of the classifier for a Gaussian input distribution is calculated empirically and shown to be significantly lower than the true VC dimension. Thus better generalisation can be predicted without the risk of over-fitting.