Dynamic Local Feature Selection in Incremental Clustering

In this paper we describe a preliminary study into the use of feature selection in incremental hierarchical clustering. Our aim is to add this capability to the clustering system, still maintaining the in-cremental nature of the learning process. This constraint lead us to consider a dynamic feature selection mechanism which is performed parallel to the clustering process. In addition, feature selection is performed in a local manner for every individual node. We propose a mechanism based upon an ordering scheme that relies on feature relevance computations. Preliminary experiments suggest that feature selection can provide a signiicant speed-up in the learning and prediction tasks without decreasing neither the quality of the obtained clus-terings nor their predictive accuracy.