A cost-sensitive ensemble classifier for breast cancer classification

Breast cancer is the most commonly diagnosed form of cancer in women. Pattern classification approaches often have difficulties with breast cancer related datasets as the available training data are typically imbalanced with many more benign cases recorded than malignant ones, leading to a bias in the classification and insufficient sensitivity. In this paper, we present an ensemble classification algorithm that addresses this problem by employing cost-sensitive decision trees as base classifiers which are trained on random feature subspaces to ensure diversity, and an evolutionary algorithm for simultaneous classifier selection and fusion. Experimental results on two different breast cancer datasets confirm our approach to work well and to provide boosted sensitivity compared to various other state-of-the-art ensembles.