Mining Association Rules Based On Particle Swarm Optimization With Application

During recent decades, the technology of data mining has made great progress, which is the result of the conflicting movement between the rapid-increasing data and the lack of information day by day. Association rule is one of the important models of data mining, which has great value in application, especially in business decision-making. Agrawal first proposed mining association rules about sets of items in customers'transaction database in 1993. The research of rules can help us to find the link among different items in transaction database, and find out customers'buying behavior mode.Particle swarm optimization (PSO) is an optimizational algorithm based on iteration, which was proposed by Kennedy and Eberhart in 1995. As a typical example of swarm intelligence, the PSO algorithm has many unique features, such as simpler principles, less parameters, and rapider convergence speed. Further more, the realization of PSO is much simple, and it has been proved that PSO is an effective global optimization method and therefore has great potential.Financial datas include real-time trading information in the process of securities, which can accurately capture the process of the changes of stock market. The association rules among these stock datas can be mined by the technology of data mining. Investors can know the rules between stock datas and stock market trends by the results of date mining, and finally make right investment decisions.We give a more systematic analysis and research on PSO algorithm in this paper, and propose a method of mining association rules based on PSO algorithm. By using the advantages of PSO, the process of data mining can be sped up, the mining efficiency can be improved, and the potential association rules in the stock market can be mined as well. The main work in this paper is as follows:1,Put forward an improved multi-swarm particle swarm optimization based on varying population size.The particles were initialized with n swarms in the algorithm, we can dynamically adjust the size of swarm by calculating the changes of Pg of each swarm. If the changes of Pg of one swarm do not occur in several generations (or changes very little), then we should reduce the number of particles into this swarm. On the contrary, if a swarm has been in a state of changes of Pg all the time, then we should increase the number of particles into this swarm. It is the same situation for all the swarms.2,Bring forward an algorithm of mining association rules based on particle swarm optimization. We use a real number coding method in this algorithm, a positive integer is represented for an attribute value, and a real number string is represented for a particle, so we can implement the algorithm through an array of operations. We also use two cooperative particle swarms in the algorithm, particle swarm of properties and particle swarm of rules. The particle swarm of properties is used to mine frequent itemsets, and the particle swarm of rules is used to mine association rules. In this way, the two stages of mining association rules are put together, and we can realize the algorithm by setting each particle swarm in a different fitness function. This algorithm needs only to scan the database once, and experiments show the algorithm is feasible, and has a better performance in improve mining efficiency, obtains a better efficiency of running time as well.3,Implement the application of mining association rules based on PSO in forecasting stock market trends.At present, the majority methods of mining stock datas have used Apriori algorithm or its improved algorithms. Although some rules can be mined, it could not avoid the inherent shortcomings of Apriori algorithms, and the numbers of mining rules are limited.The method of mining association rules based on PSO can also be used to mine the rules in the stock market. By processing original stock data, we can not only dig out the potential rules behind the transaction, but also verify the effectiveness of the algorithm, and mine more comprehensive rules.