In this dissertation, we develop a technique to cluster time series. The main idea is to use wavelets.
The time series are first expressed by means of wavelet coefficients associed to different scales. Next, a distance based of this coefficients is computed between each pair of time series. The distance matrix obtained is then used in an hierarchical algorithm. We consider several distances and algorithms.
The method is then applied to artificial and real data. These tests make us able to improve and asses the method.
|Date of Award||2004|
|Supervisor||Jean Paul Rasson (Supervisor), MARCEL REMON (Jury) & Andre Hardy (Jury)|