Outlier Resistant PCA Ensembles
Statistical re-sampling techniques have been used extensively and successfully in the machine learning approaches for generation of classifier and predictor ensembles. It has been frequently shown that combining so called unstable predictors has a stabilizing effect on and improves the performance of the prediction system generated in this way. In this paper we use the re-sampling techniques in the context of Principal Component Analysis (PCA). We show that the proposed PCA ensembles exhibit a much more robust behaviour in the presence of outliers which can seriously affect the performance of an individual PCA algorithm. The performance and characteristics of the proposed approaches are illustrated on a number of experimental studies where an individual PCA is compared to the introduced PCA ensemble.
Unable to display preview. Download preview PDF.
- Pearson, K.: On lines and planes of closest fit to systems of points in space. Philosophical Magazine 2, 559–572 (1901)Google Scholar
- Gabrys, B.: Combining neuro-fuzzy classifiers for improved generalisation and reliability. In: Proceedings the Int. Joint Conference on Neural Networks (IJCNN 2002) a part of the WCCI 2002 Congress, Honolulu, USA, pp. 2410–2415 (2002)Google Scholar
- Ruta, D., Gabrys, B.: Classifier Selection for Majority Voting. Special issue of the journal of information fusion on Diversity in Multiple Classifier Systems 6(1), 63–81 (2005)Google Scholar