基于改进流形学习的数据分类算法  

DATA CLASSIFICATION ALGORITHM BASED ON MODIFIED MANIFOLD LEARNING

作　　者：关健生[1] 

机构地区：[1]厦门理工学院电子与电气工程系,福建厦门361024

出　　处：《计算机应用与软件》2014年第12期60-63,共4页Computer Applications and Software

基　　金：福建省教育厅A类科技项目(JA12246)

摘　　要：针对高维数据分类问题的特点,提出一种基于改进型局部线性嵌入LLE(Locally Linear Embedding)算法的数据降维算法,结合支持向量机SVM(Support Vector Machine)算法实现数据分类。首先,通过LLE算法降维后的数据集,按照数据集内的离差最小化,数据集间的离差最大化的原则,计算得到最优化邻近点个数;其次,将最优邻近点个数所得的降维数据作为最优结果,按一定比例选取训练集,输入SVM算法建立数据分类器;最后,将测试集输入训练完成的分类器中,实现最优化数据分类。选取Iris flower,Yale等多类数据集与传统算法进行对比实验,验证算法的可行性。实验结果表明:所提出的算法可以有效地完成数据分类,针对低维数据和高维数据分类问题具有较好的适用性和优越性,在人脸检测中也取得较好的结果。In light of the feature of high dimensional data classification, we propose a modified LLE-based data dimensionality reduction algorithm, and implement data classification in combination with support vector machine.First we get by calculation the number of the optimised neighbouring points according to the principle of minimising the deviation within dataset and maximising the deviation between data-sets through the dataset of dimension-reduced by LLE algorithm.Secondly, we use the dimension reduction data derived from the number of optimal neighbouring points as the optimal result, choose the training sets to certain proportion and input them to SVM algorithm to set up data classifier;Finally, we input the test sets to the training-completed data classifier to implement the optimised data classification.We select the multiple-class dataset of Iris flower and Yale, etc.to carry out contrast experiments with traditional algorithms for verifying the feasibility of the algorithm.Experimental results show that the proposed algorithm can effectively implement data classification, and has better applicability and superiority for low dimension and high dimension data classification.In face detection it also obtains satisfied result.

关 键 词：数据分类 局部线性嵌入 最优邻近点个数 降维 最大化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]