Keywords: Classification, clustering, wavelets
In this paper we study several wavelet-based procedures for classification purposes. In some situations, the time domain approach may not lead to clear classification or discrimination. When we move to the wavelet domain, the multiresolution analysis leads to look at data in several levels of resolution (or scales) and then the separation may become better. One procedure uses multifractal spectra (MFS) and associated descriptors. This procedure is combined with several known classification schemes, such as classical linear and quadratic discriminant analysis, decision trees, regression methods and support vector machines (SVM). A second procedure combines discrete wavelet transform (DWT) with classification methods, in particular Schur monotone measures and classification expectation maximization (CEM). Both procedures were applied by several authors to some relevant practical problems, as classification of mammograms, SAA deficiency, high frequency pupilary responses and scale-temporal clustering in fMRI. A third procedure uses a distance computed between a given curve to be classified and all curves in the training data set. Finally, we discuss Bayes procedures for classification purposes. Simulations will be presented and also an application to a visual and auditory experiment.