Font Size: a A A

Research On The Algorithm Of Enhanced SVDD

Posted on:2019-09-16Degree:MasterType:Thesis
Country:ChinaCandidate:X ZhangFull Text:PDF
GTID:2428330548959291Subject:Computer technology
Abstract/Summary:PDF Full Text Request
The problem of one-class classification is becoming more and more extensive,which is a new branch of research in data classification.At present,many scholars have studied one-class classification problems and designed many one-class classifiers which can solve this problem.The one-class classifier is different from the two classifiers,which only describes the data of a certain class,and usually only uses the sample with a large number of samples.This paper aims to study the enhanced one-class classification algorithm based on the existing support domain method for solving one-class classification problems.The research contents of this paper include:1.Two kinds of one-class classification methods based on support domain are summarized and explained theoretically.The equivalence of these two methods under radial basis function is proved by theoretical deduction,and the research results of these two methods are summarized and analyzed.The advantages and disadvantages of different methods are used to determine the research direction of this paper.2.The support vector data description algorithm is a support domain method for solving one-class classification problems and has a high time complexity.Therefore,reducing the time complexity of this algorithm is one of the key research contents of this paper.The support vector data description algorithm uses a single type of data as a training target to train decision boundaries that are different from other classes.In order to overcome the time cost of the algorithm to solve the convex quadratic programming problem,this paper takes the divide and cure method as the main idea,describes the classification boundary as its own task,fully mines the sample distribution information,and measures the marginality of the sample by the marginality.Reduce the purpose of the training set,thereby reducing the time complexity of the original algorithm.The improved algorithm has a better datadescription performance.3.The improved support vector data description method is used to solve the problem of semi-supervised classification of unbalanced data sets.Firstly,First,the positive samples are subjected to stratified sampling preprocessing;Then,the one-class classifier is trained in the dynamic generation of different feature subspace;In the end,the untagged samples were studied by collaborative technology,and the most reliable category markers were given for unlabeled samples.This method eliminates the impact of unbalanced data sets on the classification results and effectively extends them to semi-supervised classifications.4.Collect data sets and design experiments.In the experiment of reducing the time complexity of the support vector data description algorithm,the proposed algorithm and the existing algorithms are compared using simulation data sets and real data sets.The F1 value and time-consuming are used as evaluation indicators of the algorithm.In the semi-supervised classification experiment,multiple data sets with different unbalance ratios are used for comparison experiments and the experimental results are recorded.In addition,different imbalance ratios are set for specific data sets to form multiple unbalanced data sets,through the horizontal comparison experiment,obtains the classification performance of different algorithms under a certain data set with different degrees of unbalance.Using G-mean values to evaluate the classification performance of each algorithm under imbalanced data sets.After a lot of experiments and comparisons,the classification of the algorithm in this paper is more accurate and stable,and it can well solve the practical problems in related fields.
Keywords/Search Tags:one-class classification, edge degree, time complexity, unbalanced data set, semi-supervised learning
PDF Full Text Request
Related items