The Research Of Classifiers Ensemble Based On Kernel Function And Application

Classify is the important field of pattern recognition and machine leaning. Due to the lack of standard of evaluating the classifier, so the method of increase the accuracy and generalization ability of classifier become a main research direction. In order to make full use of the mutual information of different classifiers, Suen C Y, Nadal C, Mai T A proposed the method of ensemble multiple classifiers. Ensemble classifiers improves the performance by making the full use of the mutual information of different classifiers. Ensemble classifiers is an important part of ensemble learning. By the contrast of the individual classifier, ensemble classifiers can significantly improve the performance and generalization ability of classifier by making the full use of the preference of different classifiers . Nowadays, researchers have proposed many ways of optimize and improve the ensemble classifiers, which can improve the performance of the classifier.Although ensemble classifiers can significantly improve the performance of classifier, but the accuracy of traditional ensemble classifiers will drop dramatically towards some indivisible data. First of all, this dissertation proposed a clustering algorithm based on field theory (CABFT), and proposed a CABFT based on elastic theory (CABFT with TE) through improved the CABFT with the elastic theory. After that the indivisible data with low dimensions transformed to divisible data with high dimensions by kernel function. And then the data with high dimensions clustered by CABFT. Afterward the distribution information of the data with high dimensions is obtained. At last ensemble classifiers is formed with the distribution information. The ensemble classifiers algorithm based on kernel function (CE with Kernel Function and CABFT) is proposed. Meanwhile the algorithm is improved with the theory of elasticity and the theory of marginal change. So ensemble classifiers algorithm based on kernel function with theory of elasticity (CEKF with MU) is proposed to improve the performance of classifier. Due to the lack of automatic management function of vehicle mounted strengthening 4 units iSCSI disk management system, both the algorithm of CEKF with MU and CE with KernelFunction and CABFT are applied in the vehicle mounted strengthening 4 units iSCSI disk management system. So the system can automatic manages itself. This vehicle mounted strengthening 4 units iSCSI disk intelligent management system employs the different measures with the different results of the fault data, and becomes to the intelligent management standard. And the effective automatic management mode is formed.This dissertation transforms the indivisible data with low dimensions transformed to divisible data with high dimensions by kernel function, with the theory of elasticity and the theory of marginal change, in order to improve the performance of ensemble classifiers. At last the new algorithm is applied in the vehicle mounted strengthening 4 units iSCSI disk management system in order to make the disk system become intelligent. The main contributions of this dissertation are summarized as follows:1. A clustering algorithm based on field theory (CABFT) is proposed. This algorithm improves the accuracy by the similarity of the similar objects, and the generalization ability by the difference of the different objects.2. A CABFT algorithm based on elastic theory (CABFT with TE) is proposed. According to the elastic theory, the concept of data elastic is proposed. The effect to classifier of data adjusts by the different of the data elastic. And the clustering algorithm based on field theory (CABFT) is improved. So the CABFT algorithm based on elastic theory (CABFT with TE) is proposed. Experiment show that the CABFT algorithm based on elastic theory has better clustering effect and generalization ability than the clustering algorithm based on field theory.3. The ensemble classifiers algorithm based on kernel function (CE with Kernel Function and CABFT) and ensemble classifiers algorithm based on kernel function with theory of elasticity (CEKF with MU) is proposed. First of all, the indivisible data with low dimensions transformed to divisible data with high dimensions by kernel function. And then the data with high dimensions clustered by CABFT. Afterward the distribution information of the data with high dimensions is obtained. At last ensemble classifiers is formed with the distribution information. Meanwhile the algorithm is improved with the theory of elasticity and the theory of marginal change.4. The algorithm of CEKF with MU is applied in the vehicle-mounted strengthening 4 units iSCSI disk intelligent management system. Toward the characteristic of the vehicle mounted strengthening 4 units iSCSI disk intelligent management system, analyzes the data through the algorithm of CEKF with MU. The system employs the different measures with the different results of the fault data, and becomes to the intelligent management standard. And the effective automatic management mode is formed.