A new principal component analysis method based on robust diagnosis

Classic principal component analysis is vulnerable to outliers. A new PCA method based on robust diagnosis is developed in this paper. in order to detect the outliers, a stepwise diagnosis procedure is first constructed, in which a new robust distance is defined by means of the median of observations. Observation with distance larger than a threshold will be rejected as outlier in the stepwise diagnostic procedure. Classic PCA is finally employed for the bulk of data after such a diagnosis. The efficiency of the proposed method is illustrated by several simulation and real data sets.