Unsupervised Local and Global Weighting for Feature Selection

摘要

In this paper we will describe a process for selecting relevant features in unsupervised learning paradigms using a new weighted approachs: local weight observation "OBS-SOM", and global weight observation "GObs-SOM" This new methods are based on the self organizing map (SOM) model and feature weighting. These learning algorithms provide cluster characterization by determining the feature weights within each cluster. We will describe extensive testing using a novel statistical method for unsupervised feature selection. Our approach demonstrates the efficiency and effectiveness of this method in dealing with high dimensional data for simultaneous clustering and weighting. These models are tested on a wide variety of datasets, showing a better performance for new algorithms or classical SOM algorithm. We can also show that through deferent means of visualization, OBS-SOM, and GObs-SOM algorithms provide various pieces of information that could be used in practical applications.