On the optimal weighting of high-dimensional Bayesian networks
Tatjana Pavlenko, Dietrich von Rosen · 2004
For an augmented Bayesian network classifier we propose a method of scoring a set of feature nodes for the separation strength, wherein we have combined a weighting technique and growing dimension asymptotics in a single framework. We show that the distribution of the weighted classifier is asymptotically Gaussian and establish the weight-function which is optimal in a sense of minimum misclassification probability.