KL-EEM: A Novel Kullback-Leibler Distance Based EEM Clustering Model for Breast Cancer Identification

Breast cancer is one the most common malignancies among women, along with the development of machine learning, clustering models have been utilized in this scenario. We focus on the exemplar-based clustering model in this paper. Based on the probability version of EEM clustering model, the choice of distance measurement could be more extensive. Kullbacek-Leibler distance is one of the most effective distances in Information theory, and could naturally measure the similarity between two probability distributions. Thus, this paper newly proposed a KL-distance based EEM clustering model for breast cancer by introducing KL-distance to keep target data similar to the training data into the probability version of EEM model. Deeply analyzing KL-EEM and EEM clustering models, we can conclude that the similarity based on KL-distance can be embedded into the calculation of similarity matrix. Thus, the optimization framework of EEM can be directly used to optimize the new target function of KL-EEM, which has been proved to be more effective and efficient compared with that in AP clustering model. Experiments with two real-world breast cancer datasets have been applied, and verified KL-EEM clustering model.