We present an original method for partitioning by automatic classi- fication, using the optimization technique of tabu search. The method uses a classical tabu search scheme based on transfers for the minimization of the within variance; it introduces in the tabu list the indicator of the object transfered. This method is compared with two stochastic optimization-based methods proposed by the authors (one based on simulated annealing and the other on a genetic algorithm), and with the classical k-means and Ward methods. Results of the tabu search are significantly better than the classical and genetic methods, and slightly better than our simulated annealing method.
Citation
Preprint 1-1999 CIMPA, University of Costa Rica 2060 San Jose, Costa Rica. March 1999