COMPARATIVE-ANALYSIS OF STATISTICAL PATTERN-RECOGNITION METHODS IN HIGH-DIMENSIONAL SETTINGS

An extensive simulation study is reported comparing eight statistical classification methods, focusing on problems where the number of obser vations is less than the number of variables. Using a wide range of ar tificial and real data sets, two types of classifiers are contrasted; methods that classify using all variables, and methods that first redu ce the number of dimensions to two or three. The simulations identifie d regularized discriminant analysis as the overall clearly most powerf ul classifier, and show that in most cases, a reduction of the dimensi onality to two or three dimensions prior to classification increases t he error in allocating test observations.