A solution to the classification problem with cellular automata

Abstract Classification is the task of labeling data instances by using a trained system. The data instances consist of various attributes and in order to train the system, a set of already labeled data is utilized. After the training process, success rate of the system is determined with separate test sets. Various machine learning algorithms are proposed for the solution of the problem. On the other side, Cellular Automata (CA) provide a computational model consisting of cells interacting with each other based on some predetermined rules. In this study, a new approach is proposed for the classification problem based on CA. The method maps the data instances in the training data set to cells of an automaton based on the attribute values. When a CA cell receives a data instance, this cell and its neighbors are heated based on a heat transfer function. A separate automaton is heated for each class in the data set and hence a characteristic heat map is obtained for each class at the end of the procedure. Then new instances are classified by using these heat maps. The success rate of the algorithm is compared with the results of other known classification algorithms in the experiments carried out.