Keywords: global optimization algorithm, data mining.

In modern societies, computers are used in many industries, businesses and government agencies. As a result, a lot of data has been collected during recent years. The corresponding data bases are usually designed to serve some special purpose. However, the data may sometimes provide other valuable information beyond that originally intended. Data mining was developed to serve the purpose to dig out the useful information buried in a huge data set. In recent years, data mining has been successfully applied to various areas to obtain valuable information for business, marketing, credit checks, medical research, and so forth [1,2].

In general, the embedded rules and patterns can be extracted using three different data mining activities. The first activity is called classification which determines whether a previously unknown object belongs to a known class. The second one is the association activity which finds items that have a relationship. The last one is the clustering activity which divides data into a certain number of clusters with similar behaviours. These activities in data mining are found to be very useful in helping the optimization search in this research.

When searching for the global optimum solution, the rule of thumb is that the smaller the search space, the higher the probability of finding the global optimum solution. In this research the three activities in data mining are found very suitable to play a role in reducing the search space. For unconstrained optimization problems, data mining technique is employed to generate a much smaller area where the global optimum solution may be located. For constrained optimization problems, the feasible region is identified by data mining activities. After reducing the original design space to a much smaller search space, any evolutionary algorithm or gradient-based method can be used to search for the optimum solution in this reduced space. Because the search effort is concentrated in a smaller area, the chance of finding the global optimum solution will certainly be increased.

A hybrid algorithm incorporating data mining (DM), evolution strategy (ES) [3] and sequential quadratic programming (SQP) has been developed in this study to find the global optimum solution for some benchmark test problems. The results prove that the idea proposed in this study indeed drastically increases the chance of finding the global optimum solution for some difficult test problems.

1

P. Adriaans, D. Zantinge, "Data Mining", Addison-Wesley Professional, 1996.

2

J. Han, M. Kamber, "Data Mining : Concepts and techniques", Morgan Kaufman Publishers, 2001.

3

H.-P. Schwefel, "Numerical optimization of computer models", John Wiley & sons Ltd, Chichester, U.K., 1977.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £70 +P&P)