FULL VALIDATION PROCEDURES FOR FEATURE-SELECTION IN CLASSIFICATION AND REGRESSION PROBLEMS

Those validation procedures are discussed which are commonly applied in regression and classification methods. When a chemometric technique works through several steps (scaling plus classification, feature selection plus regression, parameter optimization plus classification, etc.), validation can give an overestimate of the performance of methods if it is applied only to the final step. The idea of full validation, applied to the whole technique, is explained with reference to examples of synthetic and real data sets.