Linear classifier combination via multiple potential functions

A vital aspect of the classification based model construction process is calibration scoring function. One weaknesses that it does not take into account information about relative positions recognized objects in feature space. To alleviate this limitation, paper, we propose a novel concept calculating function on distance object from decision boundary and its to class centroid. An important property proposed score has same nature for all linear base classifiers, which means outputs these classifiers are equally represented have meaning. The approach compared with other ensemble algorithms experiments multiple Keel datasets demonstrate effectiveness our method. discuss results experiments, use performance measures statistical analysis.