Genetic Algorithm And Support Vector Machine Hybrid Method And Its Application

The growing glut of data in the worlds of science, business and government create an urgent need for a new generation of automated and intelligent tools and techniques which can analyze, summarize, and extract "knowledge" from raw data. Most knowledge discovery or data mining tools and techniques are based on statistics, machine learning, pattern recognition or artificial neural networks. The great challenge for data mining comes from huge databases of noisy, high-dimensionality data. Genetic algorithms (GAs) are good candidates for attacking this challenge since GAs are very useful for extracting patterns in high-dimensionality problems where heuristic knowledge is sparse or incomplete.The data mining approach normally includes the three major steps in the knowledge discovery process: selection, cleaning, transformation and projection of data; mining the data to extract patterns; and evaluating and interpreting the results. The first step is data preprocessing, which is important before any learning or discovery algorithms of data mining are carried out. The key operation of data preprocessing is feature selection and extraction. Mining is only one step in the overall process. The quality of mined information depends not only on the effectiveness of the data mining technique used, but also on the quality and quantity of the data preprocessed. All of these steps are usually treated as independent on the path from data to knowledge, but any one step can result in changes in preceding or succeeding steps, often requiring starting from scratch with new choices and settings.In this paper, we take regression as the main data mining task to show the general model of our approach. The data mining approach we have developed is based on a genetic algorithm which combines the preprocessing step of feature selection and extraction and the regression step into an automated loop. In the first step, GA is responsible for generating new feature space towards better direction; in the second, SVRM is used to give a nice generalization error as the value of valuation function of GA. When such a method is applied to some prediction model, excellent results have been achieved.