Classification des séries temporelles multivariées par l'usage de Mgrams (original) (raw)
Related papers
Sur l'estimation des modèles autorégressifs d'ordre multiple de séries temporelles
Comptes Rendus De L Academie Des Sciences Serie I-mathematique, 2001
Reçu le 5 mai 2000, accepté après révision le 12 février 2001) Résumé. Dans cette Note, nous définissons l'estimateur du minimum de distance d'Hellinger d'un modèle autorégressif d'ordre multiple. Sous des conditions d'existence de certaines propriétés probabilistes et d'autres hypothèses supplémentaires, nous établissons les propriétés asymptotiques de cet estimateur. En guise d'exemple, nous considérons les modèles exponentiels autorégressifs. 2001 Académie des sciences/Éditions scientifiques et médicales Elsevier SAS
Modélisation de séries temporelles multiples et multidimensionnelles
Monde des Util. Anal. Données, 2010
The most relevant elements in this paper are the automatic extraction of temporal data from Official databases and the modelization attempt of some multiple time series by exogenous other multiple time series. The results are applied on to an Epidemiological problem of modeling cancer rates incidence over twenty years, for different countries all over the world. Many issues come up when getting the data: most of the data bases are not available in the same format, some data bases are limited in terms of the number of lines that are allowed for a single query, and after importing the data, one needs to have coherence and continuity over time for each variable. The variables may cover various domains and their definition may have changed over time: expert knowledge is needed to achieve the final attribute coding and validate the retained data. A pre processing phase is then carried on: splines functions for smoothing atypical values and for filling the remaining missing data by interp...
Classification automatique de donn\'ees temporelles en classes ordonn\'ees
Cet article propose une méthode de classification automatique de données temporelles en classes ordonnées. Elle se base sur les modèles de mélange et sur un processus latent discret, qui permet d'activer successivement les classes. La classification peut s'effectuer en maximisant la vraisemblance via l'algorithme EM ou en optimisant simultanément les paramètres et la partition par l'algorithme CEM. Ces deux algorithmes peuventêtre vus comme des alternativesà l'algorithme de Fisher, qui permettent d'améliorer son temps de calcul.
This thesis is centered on the problem of the identification of time series models with the meeting of two fields of the Statistics, Time Series Analysis and Data Analysis with its descriptive methods. The first stage of our work is to extend to several discrete time series the Jenkins' principal component study developed in the Seventies. Our approach adapts "classic" Principal Component Analysis (PCA) to time series while taking as a starting point the technique Singular Spectrum Analysis (SSA). A principle is deduced and applied to the multidimensional process generating series. A Toeplitz bloc covariance matrix is built around lagged random vectors: it exploits the chronology, the stationarity and the double dimension of the process. Using two corollaries based on the tensorial product of matrices and established by Friedman B. in the Fifties, like the covariance properties of a circular process, we approach the eigenvalues and the eigenvectors of the covariance ma...
2008
In this paper, we propose an extended version of the Hilbert spectral analysis which is done in the framework of Empirical Mode Decomposition. This arbitrary order Hilbert spectral analysis is able to characterize the intermittency properties of multiscaling time series. The method is validated using simulations of fractional Brownian motion with many different values of the H parameter, and with lognormal multifractal simulations. An application is shown to real data from the field of turbulence. The method proposed here works in the amplitude-frequency space, and is the first approach able to deal with intermittency exponents directly in the frequency space. It is also shown to be superior to the classical structure functions framework to retrieve backgraound scaling fluctuations in case of a strong periodic component.
Modélisation de séries temporelles en présence d'outliers
2011
The outliers are increasingly considered in the statistical literature on time series, and this interest is increasing in econometrics. A synthesis of the rich variety of methods for treating outliers in time series stationary in relation to their nature and purposes of the investigation is the first concern of this work. The second gives a particular interest in iterative procedures for ARMA modeling. These are techniques for modeling outliers based on the test of Fox (1972) and analysis with the intervention of Box and Tiao (1975). The test gives in the same time the position and type of intervention is used as part of complete strategy for the modeling of outliers. Louni (2008) developed a test for two types of outliers (AO and IO) that seems to work well in many practical situations, especially when it is used repeatedly. Repeat the procedure of Chang et al. (1988) to identify and correct both types of outliers based on this new test and then proceed to compare the performance r...