Pattern Recognition, 4th Edition

15.11. Other Clustering Algorithms

A clustering algorithm that is based on the so-called tabu search method is presented in [Al-S 95]. Its initial state is an arbitrarily chosen clustering. The algorithm proceeds as follows. Based on the current state of the algorithm, a set of candidate clusterings, A, is created. The next state is chosen to be the “best” element of A, according to some criterion function. Certain criteria are used to prevent the algorithm from returning to recently visited states. The procedure is repeated for a prespecified number of iterations. Preliminary results reported in [Al-S 95] show that this algorithm compares favorably with the hard clustering and the simulated annealing algorithms. A recent tabu search based ...