|
Abstract:
|
Most of traditional classification methods are especially designed for low dimensional and homogeneous data. Nowadays data sets in many scientific fields are high dimensional, thus classification problem for high dimensional data becomes a difficult and challenging task. Furthermore, high dimensionality of data in the presence of outliers becomes even more serious problem. In recent years, partial least squares (PLS) method has been employed frequently as a dimension reduction tool for classification problems. However PLS is based on the sample covariance matrix, therefore, affected by outliers. In this study, we demonstrate the effects of outliers on the classification methods based on PLS and also propose a robust dimension reduction PLS method (RCPLS) for classification using a recent outlier detection method, PCOUT, especially designed for high dimensional data sets.
|
