Research And Application Of Gene Expression Programming Classification Algorithm Based On Particle Swarm Optimization

Classification of data mining is a hot area in today’s computer application technology and theoretical research. As an effective means of data analysis, it has a wide range of applications. The evolutionary computation based classification method which is formed by combining these two techniques is an important direction of the research. Gene Expression Programming (GEP) and Particle Swarm Optimization (PSO) are two new kinds of evolutionary computation method, and they search the optimal solutions separately by simulating the biological mechanism and birds foraging behavior. Taking GEP and PSO as tools, this paper studies the applications of these two evolutionary computation methods in the distance based classification approach. The main work and achievements are as follows:1. Introduces the two algorithms, GEP and PSO, including their origins, basic processes and other related concepts. Then on this basis, compares the similarities and differences of these two evolutionary computation methods.2. Elaborates the basic idea of the distance based method in detail. Proposes the NGEP-Classification, introduces a new operator, and studies the coding and decoding problems of GEP individuals. The experiment of several data sets has verified the NGEP-Classification algorithm has a stronger searching ability and better classification result. The convergence speed of NGEP-Classification algorithm slows down later in the process and becomes easy to fall into the local optimum value. Aiming at this condition, proposes the GEPSO-Classification algorithm by introducing the PSO algorithm, and solves the coding incompatibility problem in the transformation from the GEP stage to PSO stage. The experiment proves the conditions of GEPSO-Classification falling into the local optimum value have been decreased, and the classification accuracy is higher.3. In order to promote the application of the algorithm, applies GEPSO-Classification to objected-oriented remote sensing image classification, and proposes GEPSO model based objected-oriented remote sensing image classification. The experiment of London’s aviation orthogonal projection image verifies the method is feasible.