It has been shown that the dominant independent components obtained by independent component analysis (ICA) can reveal more underlying structure of the series than prin cipal component analysis (Back and Weigend 1997). To find those dominant components, all the independent components are listed in an appropriate order and then a subset of components is selected according to the order. In Back and Weigend (1997), the order is determined based on the L1 norm of each individual component without considering their interactive contribution to the observed time ser ies. In this paper, we propose to order the components according to their contribution in the mean square error (MSE) between the observed time series and the reconstruction from these components, which leads to a typical combinational optimization problem. To avoid exhaustive search, we use an sub-optimum algorithm called Testing-and-Acceptance (TnA) to find the order. Experiments on foreign exchange rates have shown that the TnA approach outperforms the L1 norm method.
[1]
Terrence J. Sejnowski,et al.
Blind separation and blind deconvolution: an information-theoretic approach
,
1995,
1995 International Conference on Acoustics, Speech, and Signal Processing.
[2]
S. Amari,et al.
Nonlinearity and Separation Capability: Further Justiication for the Ica Algorithm with a Learned Mixture of Parametric Densities
,
1997
.
[3]
Richard A. Johnson,et al.
Applied Multivariate Statistical Analysis
,
1983
.
[4]
Andrew D. Back,et al.
A First Application of Independent Component Analysis to Extracting Structure from Stock Returns
,
1997,
Int. J. Neural Syst..
[5]
Erkki Oja,et al.
Neural Networks, Principal Components, and Subspaces
,
1989,
Int. J. Neural Syst..