߬ൌߚ^ିଶ

(3.65)

R code of Bayesian BBFNN can request to the author via an email.

.28 shows the ROC curve derived from the BBFNN model

ed for the factor Xa protease cleavage data [Yang, et al., 2006].

The ROC curve of the Bayesian BBFNN model constructed for the factor Xa

eavage data. The AUC was 0.909.

he orthogonal kernel machine

N model is expressed as y = Sw+e, where the matrix S has K

The orthogonal least square algorithm is a forward kernel

procedure [Chen, et al., 1991]. The use of the orthogonal least

gorithm for BBFNN has led to a new version of BBFNN, which

as the orthogonal kernel machine (OKM) [Yang, 2005c]. At each

election step, the system incremental information content is

ed. The feature matrix is re-written as ܁ൌሺܢଵ, ܢଶ, ⋯, ܢ௄ሻ, where

ponds to ܛ௞. The elements in ܢ௞ are the mapping values of all the

n ܛ௞. It is obvious that there are correlated or there is some

cy among raw kernels ܢଵ, ܢଶ, ⋯, ܢ௄. It is expected to find new

which have no mutual correlation. In OKM, each raw kernel ܢ௞ is

med to an orthogonal kernel ( ܘ௞) to minimise the mutual