Research And Application Of Classification Algorithm Based On Cellular Automata

In recent years, Cellular Automata (CA) as a modeling tool, because of their rich features, has received a wide attention. Study of Cellular Automata in the filed of pattern classification is a new research direction. A direction of research on Cellular Automata-based classification algorithm is using Multiple Attractor Cellular Automata (MACA) to design classification algorithms, the main representative algorithm is CA-based two-classification algorithm based on genetic algorithm(GA-CA), but this algorithm is designed for two-class problem and used to solve two-class problem which is not suitable for mult-class problem; Because of the characteristics of the construction of multi-attractor cellular automata, the crossover and mutation operations of genetic algorithm makes the convergence speed low; the GA-CA algorithm split the pattern space uniform, but in some difficult problem, the sample set distribution is non-uniform, the recognition rate of the algorithm of GA-CA is always low.In this paper, the algorithm GA-CA is improved, CA-based mult-classification algorithm based on genetic algorithm (GAM-CA) is studied, which is the promotion of GA-CA. The algorithm GA-CA is improved, CA-based classification algorithm based on improved particle swarm algorithm (IPSOM-CA) is studied, which is suitable for characteristics of the construction of mult-attractor cellular automata, the speed of the convergence is improved. CA-based classification algorithm based on improved classification and regression tree (ICART-CA) is presented, which split the pattern space non-uniform, the experiment shows the algorithm is more suitable for the problem whose distribution is non-uniform.In the application, the algorithms of GAM-CA, IPSOM-CA and ICART-CA are applied to the intrusion detection. In the experiment, the data preprocessing is done and the sample set is got. Then test these three algorithms according to the test program. The experiment shows the feasibility and validity of these three algorithms.