The majority vote is the commonly method used in the predictive clustering context for assigning the class to the resulting clusters in the training phase. However, this method has limits which could influence the results quality obtained. To overcome these problems, we proposed to incorporate a local model inside every cluster to improve the predictive performance of global model. Experimental results show that the incorporation of local models allow us to obtain better results than those obtained using the majority vote; this keeping the nice descriptive aspect of the global model.