Contributions to the development of computationally efficient clustering techniques

Abstract

The concept of multilevel theory is applied in this thesis to reduce the computation and storage requirements of various existing clustering algorithms. At the first level, the data is divided arbitrarily into a number of partitions. Samples belonging to each partition are clustered separately using some clustering algorithm. Then, from each partition of the first level, representative samples-one sample per cluster - are taken to the second level for merging. This procedure is continued till the last level, which is decided based on the memory space available. This concept is used to develop:

A Multilevel Agglomerative Clustering Algorithm A Hybrid Clustering Algorithm A Multidimensional Clustering Algorithm A Two-Level Nonlinear Mapping Algorithm

These algorithms are applied to some real-world problems.