Cluster Analysis by Binary Pretopology
Cluster analysis or unsupervised classification is an important technique in exploratory data analysis when no a priori information of the distribution of the observed data is available. A new pretopological approach for unsupervised pattern classification which is based upon the pretopological concepts of adherence and pretopological closure is presented. A discrete set derived from the raw data set is used to detect the modes which correspond to regions of high local concentrations of observations by means of pretopological transformations. Results obtained on artificially generated data demonstrate the efficiency of this new approach for unsupervised pattern classification.