Study On Clustering Analysis In Data Mining Based On Simulated Annealing Algorithm

Data Mining is a new technique, which have become increasingly popular in recent years. People can apply the research result of knowledge discovery to the data process that can support the science decision. Now data mining has become a subject, which involved lots of science domain and technology especially in combining with Computational Intelligence (CI). Clustering method is one of the core techniques in data mining. It was very important in data mining process. How to choose a clustering algorithm was decided by the clustering data, aim and application. A detailed comparison which involved usual clustering algorithm in data mine was given, and a comparing analysis of usual clustering algorithm including five synthetic evaluating criterion was also given.Based on the comparing analysis and character of clustering algorithm the simulated annealing(SA)algorithms was applied to the data clustering. Simulated annealing(SA)algorithms are random search techniques based on physical annealing process, which can prevent the optimizing process into local optimization and get the global optimization. So it is widely used in solving complex optimizing problems and finding optimal solutions rapidly for difficult high-dimensional problems.In order to overcome the localization of standard SA algorithm, a synthetic improved SA algorithm was introduced. The new SA algorithm has improved the annealing process and sample process aim at making the annealing more efficiency. Then the synthetic improved SA algorithm has been applied to the clustering analysis based on the data of more than one thousand new stocks from 1992 to 2002 in the first trade day. According to the clustering analysis of initial return (IR), most new issues have great raise in the first trade day,and investors can acquire much higher initial return than the average return of the stock market. The result also shows the improved SA algorithm could be more efficiency than the standard SA algorithm without losing any accuracy of clustering and save almost 50 percent response time.