A Hybrid and Ensemble Intelligent Pattern Classification Algorithm

We introduce a novel hybrid and ensemble intelligent classifier which is an extension of ensemble classifier. Particular emphasis is put on the task of establishing the hybrid and ensemble structure of classifier depending on the principle of multi-agent structure. The hybrid and ensemble classifier include several classifiers with different types which is regarded as a set of agents. Meanwhile, every agent is composed of a set of same type´s intelligent classifiers by choosing different initialization parameters or different training set of samples. The concrete classification process contain four steps. For the unknown samples, first we obtain a set of classification results form every agent generating by all the classifiers from the agent. Second, we provide an optimization model of obtaining the associated weights of all agents. Meanwhile, the set of classification data of every agent is divided into three clustering through k-means method, further obtain three values by choosing the medians of three clustering respectively. Third, the triangular fuzzy numbers generating by all the agents are aggregated a group consensus using the known weights. Finally the consensus is compared with the pre-defined threshold. For illustration and verification purpose, a practical example is provided to analyze the developed pattern classification approach.