CMStatistics 2021: Start Registration
View Submission - CMStatistics
B0807
Title: Robust clustering with cellwise trimming Authors:  Luis Angel Garcia-Escudero - Universidad de Valladolid (Spain) [presenting]
Diego Rivera-Garcia - Banco de Mexico (Mexico)
Agustin Mayo-Iscar - Universidad de Valladolid (Spain)
Joaquin Ortega - King Abdullah University of Science and Technology (Saudi Arabia)
Abstract: It is known that a small fraction of outlying measurements can harmfully affect classical Cluster Analysis techniques. Trimming is a simple and sensible procedure to achieve robustness in statistical procedures. Some procedures have been introduced in clustering that allows trimming complete observations. However, trimming entire observations, rather than just trimming the most outlying cells, can be too extreme at sacrificing a lot of valuable information. This is especially the case when dimension increases because many observations being completely free of outlying cells are sometimes difficult to expect. In order to deal with this problem, a cellwise trimming approach based on affine subspace approximations and robust regression techniques is introduced. The approach is particularized to functional clustering problems.