Research On Clonal Selection Algorithm And Its Applications For Classification In Data Mining

In recent years, artificial immune system inspired by the immune system is becoming a research hot area in intelligent systems. Many heuristic algorithms based on immune principle have been proposed. Among them, inspired by the clonal selection model of vertebrate immune system Clonal Selection Algorithm was designed and relied on code it can realize the search independent on the problem itself. Compared with genetic algorithm, the Clonal Selection Algorithm shows a lot of useful features. It can maintain the diversity of population better while improving the convergence rate. Therefore, it can more effectively overcome such as the premature convergence and other deceptive problems difficult to solve by the genetic algorithm. The Clonal Selection Algorithm has been widely used in optimization, pattern recognition and other engineering fields, in recent years, also applied to network security, data mining and other fields.Since the birth of the most representative Clonal Selection Algorithm proposed by Castro and others, the study mainly focused on further improving the convergence rate, solution stability and accuracy of the solution, the approach involves the following aspects:(1) Around the main aspects of the algorithm: antibody coding approach, initial population generation, cloning, mutation, selection, other mechanisms of biological immune system, such as memory, forgetfulness, chaos, adaptive and so on had been introduced.(2) Based on the main operators of Clonal Selection Algorithm:Cloning, mutation, selection, New operators had been introduced to take full advantage of the useful information during the evolution, such as learning operator based on Baldwin effectsAntibody code was binary, or decimal or other way is the focus of several researchers. This researchers paid little attention to the mode of production of antibody code bit. In this paper, the iterative increase method of the bit of antibody code and the corresponding mutation strategy was proposed, in which, an antibody raised gradually increased from a bit until the entire set of coding bits. The major contributions of the dissertation are:(1) To enhance the effectiveness of the search, based on the "importance" for the solution of antibody code bit, bit-by-bit increase method was proposed in the way of the production of the antibody code.(2) To meet the bit-by-bit increase of antibody code, adaptive bit mutation was proposed to avoid the blindness of random bit mutation.(3) The Clonal Selection Algorithm introduced two strategies above was applied to extract classification rules in data mining and the corresponding classifier was constructed.The Clonal Selection Algorithm combined proposed strategies are tested and compared on some benchmark problems .the Results of experiment show the proposed strategies are effective for improving the performance of the algorithm.