Differential evolution based multiobjective optimization for biomedical entity extraction

In this paper, we propose multi-objective differential evolution (DE) based feature selection and ensemble learning techniques for biomedical entity extraction. The algorithm operates in two layers, first step of which concerns with the problem of automatic feature selection for a machine learning algorithm, namely Conditional Random Field (CRF). The solutions of the final best population provides different feature combinations. The classifiers generated with these feature representations are combined together using a multi-objective differential based ensemble technique. We evaluate the proposed algorithm for named entity (NE) extraction in biomedical text. Experiments on the benchmark setup yield recall, precision and F-measure values of 73.50%, 77.02% and 75.22%, respectively.