A Knowledge Acquisition Method of Expert System Based on Support Vector Machine

Abstract

The invention provides a method for getting expert system knowledge based on a support vector machine. A rule extracting process based on the support vector machine comprises data preprocessing, clustering of support vectors, extracting and simplifying of a super-rectangular rule, and sample identifying based on the rule. The method has the advantages that after the feature extracting and the rule simplifying are carried out, the extracted rule is more concise and is easier to explain; when a clustering and distributing matrix is calculated, only the support vectors are clustered and identified, so the calculation amount is greatly reduced; the rule extracting method is advanced, and the diagnosing and identifying rate is higher; the support vector machine is an emerging classifying technique for data mining, and has a solid theoretical foundation and a favorable generalizing performance; and the method can be used for effectively getting the rule of the expert system knowledge, and the bottleneck of dynamic getting of the expert system knowledge is broken through.

Description

technical field [0001] The invention belongs to the technical field of information processing, in particular to an expert system knowledge acquisition method based on a support vector machine. Background technique [0002] At present, knowledge acquisition based on data mining mainly obtains knowledge from some existing data through some algorithms in machine learning or mathematical statistics. Among them, association analysis, artificial neural network, rough set and decision tree are widely used in data mining. If these algorithms can be combined with current practical applications, knowledge rules can be automatically obtained from actual data, effectively breaking through the bottleneck of knowledge acquisition. It will greatly enhance the intelligent level and knowledge acquisition ability of the expert system. [0003] In recent years, as a new classification technology in data mining, support vector machine, perfect generalization theory guidance and powerful nonli...

AI Technical Summary

Problems solved by technology

However, the overlap between the elliptical rules obtained by this method is serious, and because the K-means clustering effect is too dependent on the initial value of the cluster center, it is difficult to control the number of rules and the quality of the rules by this rule extraction method ; The other is: SVM-based hyperrectangular rule extraction algorithm, first map the training samples to the high-dimensional feature space to obtain the support vectors of the samples and the optimal classification hyperplane, and then obtain the support vectors and the cluster center Based on the construction of a super rectangle, the super rectangle is finally mapped to the coordinate axis of the sample space, and the super rectangle rule is obtained

However, since the hyperrectangle rules obtained by this method use all the features of the sample, the knowledge rules obtained in this way are complex and low in comprehensibility

Images