An instances sampling approach based on cellular automata for ensemble learning

As the instances that are difficult to classify draw more attention in ensemble learning, an instances sampling method based on cellular automata is presented. The definition of the instance cellular and the neighbor region are researched, and the instance cellular kernel is designed for describing the cellular structure and cellular dynamics rules. The dynamics transfer rules of the cellular which are suit for instances sampling are investigated by combining the dynamics transfer rule of the cellular automata with the change rule of the instances distribution. The instances distribution is modified according to the transfer rule of cellular automata. An improvable ensemble algorithm is investigated by using of the sampling method base on cellular automata. The experiment results show that our ensemble method is more accurate than those obtained through the standard method.