一种采用LLE降维和贝叶斯分类的多类标学习算法.pdfVIP

Vo l. 31 No.6 第 31 誉第6 期 系统工程与电子技术 2009 年 6 月 Systems Enginecring and Electronics jun. 2009 :lt激编号: 1001-506Xωω9)06-1467-06 一种采用 LLE 降维和贝叶斯分类的多类标学习算法 李宏，谢政，向遥，关敏 (中南大学信息科学与工程学院，湖南长沙 410083) 擒 耍:多类标串敬t据斗中，的丰样平本可能』属‘于.叫…嗣俐呻 e新斤的多#集息#栋#学习算涤，古首t 先针对多9巢L 栋a敷u揭麟的特4秘正 A属属性维数高的特点，采用 LLE 算法对多类标a掀k 据的特4征正』属再性 进行降维提取能校充整描迷敛据的一组低维特4征正』鹤a 性集$然后将多 j集峰标样本1集良按所i属跑的类标边进t 行划分并采用 贝叶斯分类模型来学习各级样本集的分类特性:根据各个分类模型的判定类标，综合得到多类标样本的最终桑林 集。将该算法分别应用到自然场景阁像和.!.阁数据的多类标分类学习中，实*结果农明，该算法针对不同的多类 标报据集均能取得很好的分英效果，且相比于其他多类标算法有更高的性能。 关键诩:多类标学习 p 朴素贝叶斯分~;自然场景图像分类p 基因..据集分集 中幽分类哥哥:丁P 39 1. 4 文献标志码:A Multi唰label learning by LL配 dimension reduction and Bayesian classification LI Hong , XIE Zheng , XIANG Yao. WU Min (Schoolof Information Science and Engineering ， 。ηtral!讪th Univ. • Changsha 410083. Chína) Abstract: Sarnples of multi幽label data may belong to more than one class. so its classification problem is much rnore complicated than single-label data. A novel multi-labelleaming algorithrn is proposed. 丁he feature attributes of multi斗abel data often have high dimensions. 50 an LLE algorithm is applied to decrease the dimen- sion ìn order to extract a group of low dimensional feature attribute 8ets whìch could completely describe data. Then rnulti-Iabel sarnples are partìtioned in terms of their belonging classes. and the classification characteristics of each group are learned by using Bayesian classification rnodel. After that. the final class-label set of multi翩la­ bel samples is obtained according to the decisìon class-label of each classification model.丁he algorithm is applìed to the multi-label classification learning of both nature scene image and g