A Classification Method based on Self-adaptive Artificial Bee Colony

Evolutionary computation (EC) techniques have become popular for solving different problems because they have powerful global search ability. Artificial bee colony (ABC) is a kind of EC techniques proposed for optimization problems. In this paper, in order to take the advantage of the global search ability of ABC to solve classification problems, a basic classification model is described, and three extended classification models are proposed based on the basic model so that classification problems can be conveniently solved by ABC. In the basic classification model, a nonhomogeneous linear equation set is firstly constructed based on the training set. Then, an objective function, which can be solved by ABC, is proposed based on the equation set. This basic classification model is extensible and three extended classification models are proposed in this paper. In order to solve the classification models efficiently, a new self- adaptive artificial bee colony with symmetry initialization (SABC-SI) algorithm, which employs a symmetry initialization method and a new selection operator, is proposed. Besides, a selfadaptive search mechanism and several new candidate solution generation strategies (CSGSs) have also been developed. We conducted experiments on eight datasets chosen from the UCI Machine Learning Repository. The experimental results show that SABC-SI can be directly used for classification by solving the classification models, and achieve good classification accuracy.