Dissertation > Excellent graduate degree dissertation topics show

Research and Application of Imbalance Data Classification Based on Support Vector Machine

Author: ZhaoWenJuan
Tutor: HanFengQing
School: Chongqing University of Technology
Course: Applied Computer Technology
Keywords: Support Vector Machine Unbalanced data Re - division of the training set Compressed convex hull Information gene
CLC: TP274
Type: Master's thesis
Year: 2011
Downloads: 73
Quote: 0
Read: Download Dissertation

Abstract

People in the information age requires processing large amounts of data to find its own rules and its use. Classification is often to do work in data processing, classification problems become an important domain of machine learning research. Support Vector Machine training set by the kernel function mapped to a higher dimensional space, it can achieve a limited sample linear and non-linear case classification. Studies have shown that support vector machines to balance the data better classification results, but poor the imbalance data classification effect, this is because of the decision of the support vector machine classification hyperplane by support vector, the majority of samples number of support vectors more than the minority class sample separating hyperplane offset, this would lead to lower recognition rate of minority class samples, found time even when the sample is a serious imbalance in the minority class samples corresponding classification rules. The main research objective of this paper is how to use the support vector machine method of unbalanced data classification, major work with innovations include the following aspects: First, the support vector machine theory. The limitations of the analysis of empirical risk minimization, introduced structural risk minimization principle and its advantages, detailed summary of the theory and research status of support vector machine. Unbalanced data classification method. The problems faced by the analysis of unbalanced data classification, focus on a variety of unbalanced data classification method to classify and summarize, and analyze the advantages and disadvantages of the various methods. Given based clustering data set into support vector machine method DISVM. The main idea is the majority class samples divided into a series of sub-sets, each divided subset, and a few sample combinations were training with SVM method, the final integration of each sub-classifier. The main methods for the previous algorithm during data set is divided without considering the subset of the shortcomings of divided rule improvements and through experiments to prove the validity of the method on the problem of unbalanced data classification. Given a the imbalance data compression convex hull support vector machine classifier method GSVM. Analysis of the geometric characteristics of the support vector machine, the first two types of samples compressed to its direction of the center of gravity, and then seek two types of samples compressed convex hull closest pair and classification hyper plane generated by support vector machine method. Experiments show that the method has good classification performance. Characteristic imbalance is also an important aspect of the unbalanced data classification, this paper mainly combined with acute leukemia gene expression data published by Golub et resolve characterized imbalance. The previous method only consider single gene disease category judgment, this article is mainly to consider the correlation of the two genes, and as a measure to screen candidate genes, and the effectiveness of the method is verified by experiments.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Automation technology and equipment > Automation systems > Data processing, data processing system
© 2012 www.DissertationTopic.Net Mobile