Two-Level Automatic Classification applied to Bulky Data Bases

Classifying the data of a bulky base on the basis of high number of attributes is not an easy task because of the scarcity of the adequate methods present in the literature. These methods generally resort to the reduction of the number of data using sampling techniques or the analysis in principal components (APC). Problems are often encountered, namely the complexity of calculation, the slowness of execution and the relevance of the results. We developed for this purpose, an approach of two-level automatic classification allowing to transform a bulky base into an exploitable group of classes for the extraction of knowledge and decision-making. The robustness, the precision and the optimality of our approach are shown through its comparison with the traditional approach of classification (classification of the original data base), and this, through the results produced following the application of two approaches to a bulky data base. These results include both the clusters and the knowledge map formed by association rules generated on the original base on the one hand, and the summary of BIRCH on the other hand