AbstractThis dissertation attempts to comprehend the conception of a new modelling process of times series' based on the use of wavelets. The series are first expressed by means of wavelets coefficient. Those are associated with different scales and with the help of several algorithms of dyadic transform. Different denoising techniques will then be applied to eventually model the wavelet coefficients on basis of symbolic data coupled with an interval mode. Afterwards, different methods relying on ascending hierarchical classifications will be approached in order to compare the results obtained for the new modelisation method of times series' classification rate with other results that we get with other classical methods handling denoising based on the Normal or Poissonning model. In the end, after having experimented the semi-artificial data, we will deal with the results that we have gathered for experimenting on real data in a contrastive approach.
|Date of Award||2005|
|Supervisor||Jean Paul Rasson (Supervisor), Andre Hardy (Jury), MARCEL REMON (Jury) & Séverine Guilmot Adans (Jury)|
Le débruitage des ondelettes dans la classification des séries chronologiques
Lippert, A. (Author). 2005
Student thesis: Master types › Master in Mathematics