Detection and visualisation of outliers using kernel principal components

We apply, in this article, a new method to identify outliers from a dataset. It consists to use the K-means clustering algorithm on the smallest principal components provided by the kernel principal components analysis. Two leading methods commonly used in the domain namely the standard deviation and the Tukey boxplot are tested and compared to our method. The experiments on artificial and real datasets show that our approach better detects outliers than the two classical methods.