Support Vector Classifier Machine Based On Feature Analysis

As a new method in the field of data mining, SVM developed as a successful approach to treat classification and regressive problems, has become a hot issue in the field of data mining. Data pre-processing is indispensable in any case of data mining for its role in redundancy eliminating, data scale reducing and computational effort saving. Feature analysis, an important step in data pre-processing, is a decisive action in data mining, for features are the main records of data mining cases and lay direct influence on data mining. Feature Selection (FS), the discretization of continuous features and Feature Weighted (FW) are important methods for feature analysis. In this thesis we try to find good feature analysis methods and combine the merits of SVMs. The main results in this thesis consist of three parts.1. Two feature selection methods are proposed: one is Feature Selection based on Classification Accuracy of SVM and the other is Floating Search Feature Selection. The two feature selection methods are effective in application.2. A discretization algorithm based on NCL clustering and CAIR criterion is proposed. Based on NCL, it mines the data well by firstly finding the initial partition points and then re-selecting partition points by using the CAIR criterion.3. It articulates the mathematical principle for feature weighted SVM and improves a weighting method.