Combining Linear Classifiers Using Score Function Based on Distance to Decision Boundary

In this work, we addressed the issue of combining linear classifiers in the geometrical space. In other words, it means that linear classifiers are combined via the combination of their decision hyperplanes. For this purpose, an approach based on the potential functions is proposed. The potential function spans a potential field around each decision plane. The potential fields coming from decision planes are superposed, and the resultant decision field is used as an aggregated model of the ensemble classifier. During the experimental study, the proposed approach was applied to an ensemble built on heterogeneous base classifiers, and it was compared to two reference methods - majority voting and soft voting, respectively. The result shows that the proposed method can improve classification performance metrics compared to soft voting.