Research On Classification System Based On Improved Genetic Algorithm

Data classification is the core of the problem in the domain of data mining. Genetic algorithm has been applied widely as a classification algorithm in data mining. As the first machine learning system, the classification system based on genetic algorithm proposed by Holland has received much concern since it was put forward. At present, it is still hot spot of the researching in data mining.In this thesis, it is found that the remarkable character of genetic algorithm is the cryptic parallelism and the effective utilizing of the global information as an overall random searching algorithm. But it can do nothing for the feedback information. Many redundant interactions are done in later stage. The low efficiency in accurate solving, poor ability in local searching and easy emergence of premature convergence is existed. In addition, the accuracy of classification in the classification system based on simple genetic algorithm isn't high enough.Aim at the problems above, the improved strategies of genetic classification algorithm in the classification system is researched in this thesis. At first, the solution space of the optimization problem is divided evenly and each subspace is marked by ant colony pheromone which controls the selection operation. Secondly, double selection operator, crossover operator based on "heterosis" and adaptive mutation operator is designed. Thirdly, the improved genetic algorithm is applied in classification system and the model of system is established. At last, the accuracy of classification, the time of the algorithm running and the convergence of algorithm is tested through the analysid of the data of the result in experiment. It is proved that the performance of the classification based on improved genetic algorithm is improved on the three aspects above.