Cluster Analysis Based on the Central Tendency Deviation Principle

Julien Ah-Pine (1)

1 Xerox Research Centre Europe [Meylan]

Our main goal is to introduce three clustering functions based on the central tendency deviation principle. According to this approach , we consider to cluster two objects together providing that their similarity is above a threshold. However, how to set this threshold ? This paper gives some insights regarding this issue by extending some clustering functions designed for categorical data to the more general case of real continuous data. In order to approximately solve the corresponding clustering problems, we also propose a clustering algorithm. The latter has a linear complexity in the number of objects and doesn't require a pre-defined number of clusters. Then, our secondary purpose is to introduce a new experimental protocol for comparing different clustering techniques. Our approach uses four evaluation criteria and an aggrega-tion rule for combining the latter. Finally, using fifteen data-sets and this experimental protocol, we show the benefits of the introduced cluster analysis methods.

Mots clés: Cluster analysis; clustering functions; clustering algorithm; categorical and real continuous data; experimental protocol

Domaines: Apprentissage [cs.LG]; Machine Learning [stat.ML]

Julien Ah-Pine. Cluster Analysis Based on the Central Tendency Deviation Principle. 5th International Conference on Advanced Data Mining and Applications (ADMA 2009), Aug 2009, Pékin, China. pp.5-18, ⟨10.1007/978-3-642-03348-3_5⟩. ⟨hal-01504464⟩