The performance of k-means clustering method based on robust principal components

The k-means clustering method is the most widely used method to group n observations into k clusters. It is now evident that clustering results can be improved by applying classical principal component analysis (PCA) with the k-means clustering algorithm. However, the clustering results of PCA with k-means are adversely affected by the presence of outliers in a data set. To remedy this problem, we proposed to integrate robust principal component analysis (RPCA) with the k-means algorithm. Simulation study and real examples are carried out to compare the performance of the classical k-means, k-means based on PCA and k-means based on RPCA. The findings indicate that the k-means based on RPCA outperforms the other two methods.

Actions (login required)

View Item