|
Abstract
|
|
|---|---|
| In this paper it is presented a study about the convenience of applying a wavelet multiresolution analysis to analyze and forecast a time series based on the Hurst exponent calculation. It is also presented the direct application to complex neural networks classification stages design. The Hurst exponent analysis gives an approximation of the predictability of a time series, so this point gives the key information to understand if a time series can be analyzed in a classical analysis-synthesis wavelet analysis and the optimum decomposition degree level. This criterion can be directly translated in the feature selection stage in a Neural Classifier design. A rainfall time series is studied as a case study performing two different wavelet analysis and selecting the best one in terms of the Hurst¿s Exponent | |
|
International
|
Si |
|
Congress
|
1st European Workshop on Turbulence and Fractals |
|
|
960 |
|
Place
|
Madrid, España |
|
Reviewers
|
Si |
|
ISBN/ISSN
|
1870-4069 |
|
|
|
|
Start Date
|
10/12/2009 |
|
End Date
|
10/12/2009 |
|
From page
|
1 |
|
To page
|
5 |
|
|
Proc. of 1st European Workshop on Turbulence and Fractals |