Miscellanea
Ludmila I. Kuncheva · 2004
Chapter 8 contains three shorter topics: feature selection, error correcting output codes (ECOC) and cluster ensembles. Methods for feature selection for classifier ensembles are described including random selection (pure random selection and genetic algorithms) and non-random selection. Code designs for ECOC ensembles are presented including exhaustive, one-per-class, and random codes. The last part of the chapter introduces combining clustering results or cluster ensembles. We discuss measures of similarity between partitions and approaches to evaluating clustering algorithms. Two combiners for cluster ensembles are included: majority vote and direct optimization of a matching function between the individual outputs and the ensemble result.