English

Outliers Detection Is Not So Hard: Approximation Algorithms for Robust Clustering Problems Using Local Search Techniques

Data Structures and Algorithms 2021-01-01 v2 Combinatorics

Abstract

In this paper, we consider two types of robust models of the kk-median/kk-means problems: the outlier-version (kk-MedO/kk-MeaO) and the penalty-version (kk-MedP/kk-MeaP), in which we can mark some points as outliers and discard them. In kk-MedO/kk-MeaO, the number of outliers is bounded by a given integer. In kk-MedP/kk-MeaP, we do not bound the number of outliers, but each outlier will incur a penalty cost. We develop a new technique to analyze the approximation ratio of local search algorithms for these two problems by introducing an adapted cluster that can capture useful information about outliers in the local and the global optimal solution. For kk-MeaP, we improve the best known approximation ratio based on local search from 25+ε25+\varepsilon to 9+ε9+\varepsilon. For kk-MedP, we obtain the best known approximation ratio. For kk-MedO/kk-MeaO, there exists only two bi-criteria approximation algorithms based on local search. One violates the outlier constraint (the constraint on the number of outliers), while the other violates the cardinality constraint (the constraint on the number of clusters). We consider the former algorithm and improve its approximation ratios from 17+ε17+\varepsilon to 3+ε3+\varepsilon for kk-MedO, and from 274+ε274+\varepsilon to 9+ε9+\varepsilon for kk-MeaO.

Keywords

Cite

@article{arxiv.2012.10884,
  title  = {Outliers Detection Is Not So Hard: Approximation Algorithms for Robust Clustering Problems Using Local Search Techniques},
  author = {Yishui Wang and Rolf H. Möhring and Chenchen Wu and Dachuan Xu and Dongmei Zhang},
  journal= {arXiv preprint arXiv:2012.10884},
  year   = {2021}
}
R2 v1 2026-06-23T21:06:24.388Z