A Family of GEP-Induced Ensemble Classifiers

The paper proposes applying Gene Expression Programming (GEP) to induce ensemble classifiers. Four algorithms inducing such classifiers are proposed. The first one, denoted GEPA, based on the Adaboost method, is the two-class specific. The second, denoted MV is based on majority voting learning. Third one, denoted MVI, assumes incremental learning where for some classes more genes may be needed than for other ones. Finally, the last one denoted MVC involves partitioning of the training dataset into clusters prior to expression trees induction. The proposed algorithms were validated experimentally using several datasets.