Non parametric and robust statistical test based on wavelets for time series classification

In this article a statistical procedure for identifying if a time series set follows the same model is developed -- With the aim of supporting characterization and pattern recognition for temporal series, and inspired by the methodology of Maharaj E. A.[1], we take advantage of the wavelet coefficients properties to characterize a signal and our procedure is made by means of a randomization test on those coefficient -- Our main contribution in this work is to introduce modified versions of test statistic in test for pattern recognition of time series which in general, have a great performance in terms of size and power, both being desirable features in a statistic test -- It is worth pointing out that we introduce robust statistical tests whose performance are better in presence of atypical values than some techniques already studied in the literature -- The methodology developed here allow us to design a new method to classify time series and atypical values identification -- We implement our new methods in real and simulated case