Automatic Clustering Algorithm Based On Artificial Immune System And Its Application

With the development of information technology, data mining is becoming one of the most advanced and active research topics in the field of the information decision-making. As an effective tool of data mining, cluster analysis has been attracting wide attention. Many of the real data information are very complex, and the number of classes is very difficult to obtain, in this case, the automatic clustering algorithm emerged. Clustering can be seen as an optimization problem, which means we can use different optimization methods to solve it.Based on the above background, this paper proposes a single-objective optimization based immune automatic clustering algorithm and multi-objective optimization based immune automatic clustering algorithm. In addition, we also made a thorough research on multi-objective optimization algorithms in dynamic environment, and propose a new dynamic multi-objective optimization algorithm. The specific work is arranged as follows:1. A dynamic local search based immune automatic clustering algorithm is proposed. In the proposed algorithm, fistly, for the structure of chromosomes, we propose a clustering center based dynamic local search to find the optimal number of categories of data sets. Secondly, the algorithm uses adaptive strategy differential crossover based on the neighborhood structure to further improve the performance of the algorithm. Through testing thirty datasets with the different distribution characteristics, four synthetic texture images and four SAR images, fexperimental results show that the proposed algorithm has a significant advantage than other four compared algorithm.2. A new dynamic multi-objective immune optimization algorithm based on prediction strategy is proposed. In the proposed algorithm, fistly, when a change in the objective space is detected, based on the non-dominated antibodies in previous optimum locations, a modified forecasting model is used to generate the initial antibodies population. Secondly, in order to avoid convergence to local optimum algorithm and further accelerate the convergence speed, an improved differential evolution crossover with two selection strategies is proposed. Experiments show that the proposed algorithm has a strong ability to track dynamic changes, the best convergence and diversity. 3. A Synergy of Two Mutation based Immune Multi-objecitve Automatic Clustering Algorithm is proposed. In the proposed algorithm, fistly, two new mutation operators, which are designed for the different structure of chromosome respectively, are cooperated with each others to generate the new choromosome. Secondly, In order to avoid the shortcoming of the PBM validity index in single-objective automatic clustering analysis, we proposed a exponential function based validity index, and combined it with PBM as two optimition objectives. over a test suit of24datasets and5synthetic texture images, experiments show that the proposed algorithm has a greater advantage in data clustering and image segmentation, and the robustness of the algorithms is best.