Skip to Main Content
A strategy is introduced to rank and select principal component transform (PCT) and discrete cosine transform (DCT) transform coefficient features to overcome the curse of dimensionality frequently encountered in implementing multivariate signal classifiers due to small sample sizes. The criteria considered for ranking include the magnitude, variance, interclass separation, and classification accuracies of the individual features. The feature ranking and selection strategy is applied to overcome the dimensionality problem, which often plagues the implementation and evaluation of practical Gaussian signal classifiers. The applications of the resulting PCT- and DCT-Gaussian signal classification strategies are demonstrated by classifying single channel tongue movement ear pressure signals and multichannel event related potentials. Through these experiments, it is shown that the dimension of the feature space can be decreased quite significantly by means of the feature ranking and selection strategy. The ranking strategy not only facilitates overcoming the dimensionality curse for multivariate classifier implementation but also provides a means to further select, out of a rank ordered set, a smaller set of features that give the best classification accuracies. Results show that the PCT- and DCT-Gaussian classifiers yield higher classification accuracies than those reported in previous classification studies on the same signal sets. Among the combinations of the two transforms and four feature selection criteria, the PCT-Gaussian classifiers using the maximum magnitude and maximum variance selection criteria gave the best classification accuracies across the two sets of classification experiments. Most noteworthy is the fact that the multivariate Gaussian signal classifiers developed in this paper can be implemented without having to collect a prohibitively large number of training signals simply to satisfy the dimensionality conditions. Consequently, the classif- - ication strategies can be beneficial for designing personalized human-machine interface signal classifiers for individuals from whom only a limited number of training signals can reliably be collected due to severe disabilities.