Research On Active Learning Support Vector Machine And Its Application

In pattern classification, According to the different study style, learning algorithm is divided into supervised and unsupervised learning. Classification accuracy of unsupervised learning algorithm is usually not satisfactory, so people tend to make use of supervised learning in practice. Support Vector Machine (SVM) is a common supervised learning algorithm. Classification accuracy of SVM algorithm is high, but it needs many appropriate labeled training samples that are sometimes difficultly gained in true-life. How to make full use of unlabeled samples which contain lots of connotative information are on focus. According to such situation, the paper combines SVM with active learning, and presents a couple of improved classification method of active learning support vector machine (ASVM). The main contributions of this thesis are given as follows:(1) We understand the future trend and direction of active learning and support vector machine, and deeply study the algorithm theory of active learning and support vector machine.(2) We introduce a simple algorithm of ASVM based on distance and an algorithm of passive learning support vector machine, then compare and analyze their classification results and verify the advantages and validity of ASVM algorithm.(3) We deeply study the algorithm of ASVM, the training samples of SVM need to be chosen by manual, and the classifier is usually affected by isolated points of samples. In order to make up for these shortages, this paper presents a new classification method of ASVM by using K-means algorithm and improving active learning. Experimental results for Iris data, Wine data and Remote sensing data verify the validity of our method.(4) While the methods of ASVM are focus on the samples which are close to the current separating hyper-plane, and its ignore some SV samples which are far form the separating hyper-plane, also its pay not attention on if the current separating hyper-plane is close to the optimal one. We propose an improved algorithm of active learning support vector machine based on probability, and the advantages and validity of new method are shown by Iris data, Wine data and Remote sensing image experiments.