Cellular Automata Based Binary Classification

In this study, nonlinear cellular automata were used for binary pattern classification. Cellular automata were first proposed by Von Neumann, who determined the working principles of today's computer architecture, to model the self-renewal abilities of biological beings. Cellular automata have a computational model based on the state update logic due to the interaction with the cells around the cells in the grid plane. Studies on cellular automata have shown that some states are in a dynamic interaction with other states. These states gathered other situations around itself and acted in the center of attraction. States that behave in the form of attraction center are called attractor state(or attractor basin). The dynamic behaviours of the attractor were considered as a pattern of attracting other patterns and revealed the potential of using cellular automata in pattern recognition and classification. The first pattern recognition methods based on cellular automata are linear methods that use rules that update the state according to XOR and XNOR logic. Later nonlinear methods have been developed to overcome the limitations of linear methods. In this study, reachability tree based nonlinear methods are used to characterize the attractors. Attractor states are used for binary classification purposes on different data sets. The results obtained were compared with previous cellular automata based pattern recognition methods and other known methods