A General Partition Cluster Algorithm
Daniel Peña, Julio Rodríguez, George C. Tiao · 2004
A new cluster algorithm based on the SAR procedure proposed by Peña and Tiao [ 9 ] is presented. The method splits the data into more homogeneous groups by putting together observations which have the same sensitivity to the deletion of extreme points in the sample. As the sample is always split by this method the second stage is to check if observations outside each group can be recombined one by one into the groups by using the distance implied by the model. The performance of this algorithm is compared to some well known cluster methods.