Title :
Estimating outlier impact on FastICA using fuzzy inference
Author :
Gadhok, N. ; Kinsner, W.
Author_Institution :
Dept. of Electr. & Comput. Eng., Manitoba Univ., Winnipeg, Man., Canada
Abstract :
The impact of outliers on signal separation performance of an independent component analysis (ICA) algorithm is an important factor when selecting an ICA algorithm. If an ICA estimator has the property of B-robustness, the influence of an extreme point is bounded, leading to good separation performance in the presence of outliers. Since this property is binary, it does not give the degree of influence an outlier has on the separation performance. To address this issue, a Mamdani-type fuzzy inference, based on the location of a potential outlier and on the skewness of the data set, has been developed. It creates an outlier sensitivity map for an ICA algorithm. The implication of this work is a criterion to switch from one ICA algorithm to another in real time, as determined by the algorithms sensitivity to the data set under consideration. This paper describes estimation of the outlier impact on the separation performance of the non-B-robust FastICA algorithm using a Mamdani-type fuzzy inference. In simulations with data sets contaminated by outliers, the FastICA sensitivity map resembles the separation performance measured by the Amari performance index.
Keywords :
fuzzy logic; independent component analysis; inference mechanisms; performance index; sensitivity analysis; source separation; FastICA sensitivity map; fuzzy inference; independent component analysis algorithm; outlier impact estimation; outlier sensitivity map; potential outlier; Blind source separation; Higher order statistics; Independent component analysis; Inference algorithms; Jitter; Pollution measurement; Robustness; Signal processing algorithms; Statistical analysis; Switches;
Conference_Titel :
Fuzzy Information, 2004. Processing NAFIPS '04. IEEE Annual Meeting of the
Print_ISBN :
0-7803-8376-1
DOI :
10.1109/NAFIPS.2004.1337410