Complete validation for classification and class modeling procedures with selection of variables and/or with additional computed variables

摘要: Abstract The evaluation of the predictive ability a model, is an essential moment all chemometrical techniques. So it must be performed very carefully. However, in case selection relevant variables (an step data sets with many, frequently thousands, variables) generally using available objects. In some recent classification and class modeling techniques, from original or selected Mahalanobis distances leverages centroids categories problem are computed, then added to variables. Also here computed consequence overestimate prediction ability, large when ratio between number objects that rather low, so variance-covariance matrix unstable. this paper correct validation procedures described for cases addition on estimates compared those obtained insufficient strategies.