Cluster Ensemble Based on Particle Swarm Optimization

Motivated by the success of ensemble methods in supervised learning problem, cluster ensembles have started to gain an increasing interest. Since the absence of labeled training data, cluster ensemble is a more challenging task than multiple classifiers system. In this paper, a novel weighted combination model of multiple partitions was proposed, in which particle swarm optimization algorithm was used to optimize the parameter. Adopting error rate as the fitness function, weights vector was treated as a particle in the search space. Then the task is converted into an optimization problem for minimum. An experimental investigation was performed on UCI repository and encouraging results were obtained. The proposed combination model obtained the best performance on 8 out of 9 data sets. On the remained one data, which is relatively smaller, its performance is between the best individual and the worst. Over fitting might be the answer for the worse behavior on smaller data set. So the proposed model is superior given larger data sets.