Decoupling of clustering and classification steps in a cluster-based classification

The application of cluster analysis in the "classification" area is well known. Such application takes place in two steps: "clustering" and "classification". In the clustering step, the objects of a training set are clustered using a cluster technique, Q. The outcome is a set of clusters, C. Each cluster, ci, is assigned a class label, ki, which reflects the common features of the objects in ci. The ki is a member of set K. In the classification step, a new object from a test set is assigned to one of the clusters in C using the Q, C, and K of the former step. The goal of this research effort is two fold: (1) introducing a methodology for decoupling "clustering" and "classification " steps and (2) establishing the validity of the proposed methodology by comparing its classification performance with the performance of the rough sets approach, and disciminant analysis.