Application of the sample-conditioned MSE to non-linear classification and censored sampling

Phenotype discrimination problems in biomedicine typically classify between types of pathology, stages of disease, response to treatment or survivability. In contrast to the usual heuristic classifier and error estimate computed from small sample data, recent work proposes a Bayesian modeling framework over an uncertainty class of feature-label distributions, which when combined with data facilitates optimal MMSE error estimation, optimal classifier design and a sample-conditioned MSE for error estimation analysis, all relative to uncertainty in the underlying distributions conditioned on the sample. Here we address application of the conditional MSE to non-linear classifiers and present an example with optimal Bayesian classification and censored sampling, an economical sampling procedure in which data are collected incrementally until desired criteria are met.