A new clustering algorithm based on the concept of graph connectivity is introduced. The idea is to develop a meaningful graph representation for data, where each resulting sub-graph corresponds to a cluster with highly similar objects connected by edge. The proposed algorithm has a fairly strong theoretical basis that supports its originality and computational efficiency. Further, some useful guidelines are provided so that the algorithm can be tuned to optimize the well-designed quality indices. Numerical evidences show that the proposed algorithm can provide a very good clustering accuracy for a number of benchmark data and has a relatively low computational complexity compared to some sophisticated clustering methods.
Advances in Intelligent Systems and Computing, Springer, pp.442-454