Semi-supervised Artificial Immune Mixture Models Clustering

Learning with partly labeled data aims at combining labeled and unlabeled data in order to boost the accuracy of an algorithm. The traditional Expectation maximization (EM) algorithm only produces locally optimal solutions, it is sensitive to initialization, and the number of components of mixture model must be known in advance. We propose a novel semi-supervised clustering algorithm that uses Gaussian mixture models (GMM) as the underlying clustering model. A novel adaptive global search mechanism is introduced into semi-supervised gaussian mixture model-based clustering, where the EM algorithm is incorporated with the ideas of an immune clonal selection technique. The new algorithm overcomes the various problems associated with the traditional EM algorithm. And it can improve the effectiveness in estimating the parameters and determining the optimal number of clusters automatically. The experimental results illustrate the proposed algorithm provides significantly better clustering results, when compared with other methods of incorporating equivalence constraints.