Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/43895
標題: An incremental support vector machine-trained TS-type fuzzy system for online classification problems
作者: Wei-YuanCheng
Chia-FengJuang
關鍵字: Fuzzy neural networks;Fuzzy classifiers;Incremental support vector machines;Channel equalization;Skin color segmentation
出版社: Elsevier B.V.
Project: FUZZY SETS AND SYSTEMS, Volume 163, Issue 1, Page(s) 24-44.
摘要: 
This paper proposes an incremental support vector machine-trained TS-type fuzzy classifier (ISVM-FC). The ISVM-FC is a fuzzy system that consists of Takagi-Sugeno (TS)-type fuzzy rules. Structure and parameters in the ISVM-FC are trained incrementally from one subset of training data at a time. This incremental training approach avoids the use of large amounts of memory required for storing training data in batch learning, reduces training time, and adapts the classifier to time-dependent classification systems where training data are available sequentially. Initially, there are no fuzzy rules for structure learning with the ISVM-FC. It generates all rules according to the distribution of the training data. An incremental linear support vector machine (SVM) is used to tune the resulting rule parameters to give the classifier better generalization performance. The use of incremental learning discards past training data adaptively according to its distance to the linear hyperplane, thereby improving learning efficiency. Three simulations are conducted to verify the performance of the ISVM-FC. Comparisons with fuzzy classifiers and Gaussian-kernel SVM with batch and incremental learning modes demonstrate that the ISVM-FC improves training and test times, and reduces memory consumption for classifier storage without deteriorating the generalization ability. (C) 2010 Elsevier B.V. All rights reserved.
URI: http://hdl.handle.net/11455/43895
ISSN: 0165-0114
DOI: 10.1016/j.fss.2010.08.006
Appears in Collections:電機工程學系所

Show full item record
 

Google ScholarTM

Check

Altmetric

Altmetric


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.