A classifier based on normalized maximum likelihood model for classes of Boolean regression models

Boolean regression models are useful tools for various applications in nonlinear filtering, nonlinear prediction, classification and clustering. We discuss here the socalled normalized maximum likelihood (NML) models for these classes of models. Examples of discrimination of cancer types by using the universal NML model for the Boolean regression models indicate its ability to select sets of feature genes discriminating at error rates significantly smaller than those of other discrimination methods.