Research On Genetic Algorithm And Its Application In Classification Rules Mining

With the fast development of internet technology, the growth rate of information increases exponentially, the traditional method is far from meeting the real needs. In order to find valuable information from huge amounts of data, data mining technology generated for meeting the demand. Classification rules mining is an important branch of data mining, many researchers used basic genetic algorithm for classification rules mining, there is always premature convergence problem. Therefore, how to improve the premature convergence of genetic algorithm to improve the efficiency of classification rules mining is a valuable research topic.Maintain the diversity of population is the key of resolving the problem of premature convergence. In order to solve these problems effectively and improve the efficiency and accuracy of classification rules mining, we improve basic genetic algorithm and put forward a kind of improved genetic algorithm based on double population and double mutation, and apply them for classification rules mining in this paper. The main work of this paper is as follows:Firstly, this paper improves the three operators of genetic algorithm. The improved select operator sorts the individuals by the fitness in ascending order, and the front two thirds of the individuals are selected by equal probability and the one third of the individuals are selected by the proportion of individual. It can maintain the diversity of population. According to the result that individual similarity compares with the set threshold, the improved crossover operator implements single-point crossover or uniform crossover by certain probability, it can improve the search capability of algorithm. Improved mutation operator makes the probability of individual in group adjust adaptively and ensures that the mutation probability of the outstanding individuals is not zero, it can maintain the population evolving and accelerate the convergence. Experimental results show that the improved algorithm maintains the diversity of population and improves premature phenomenon obviously, and excavates classification rules of better quality.Secondly, this paper puts forward a kind of improved genetic algorithm based on double population and double mutation. The algorithm has two initial population, when the initial population is produced, it is sorted by the fitness, half of the outstanding individuals will be placed in group two, and the other half will be placed in group one. Then, the individuals in group one are implemented with double mutation and the individuals in group two are implemented with single mutation. After performing a genetic manipulation, if the individual in group one is better than the individual in group two, the individual in group two will be replaced with the individual in group one. Experimental results show that the algorithm eliminates local optimum problems of basic genetic algorithm for classification rules mining and is capable of acquiring more accurate classification rules fastly.