Journal
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
Volume 50, Issue 12, Pages 5162-5171Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TSMC.2018.2867238
Keywords
Data models; Robustness; Support vector machines; Collaboration; Collaborative work; Manufacturing; Noise measurement; Cluster; collaborative learning; least squares support vector machine (LS-SVM); relative density degree; robustness
Funding
- National Natural Science Foundation of China [51675539]
- Project of Innovation-Driven Plan in Central South University [2015CX002]
- Project of State Key Laboratory of High Performance Complex Manufacturing [ZZYJKT2018-17]
- Hunan Province Science and Technology Plan [2016RS2015]
- Central South University Graduate Scientific Research Innovation Project [2018zzts452]
Ask authors/readers for more resources
The least squares support vector machine (LS-SVM) is often employed to model data with a nonlinear distribution using a divide-and-conquer strategy. However, when nonlinear data are contaminated by either noise or outliers, LS-SVM is often an ineffective approach due to a lack of robustness. In this paper, a collaborative learning-based clustered LS-SVM method is proposed for modeling of nonlinear processes that are subject to noise or outliers. First, a large-scale dataset is divided into several subsets and the data distribution of each subset is estimated. A robust LS-SVM is then developed to represent each subset using this distributional information. A global model is further constructed through integration of all submodels, whose continuity and smoothness are ensured by the development of the collaborative learning technique. As a result, the proposed method considers both the nonlinear distribution of data and the robustness of each submodel, and ensures the continuity and smoothness of the global model. Thus, it can effectively model nonlinear data that is subject to either noise or outliers. As further validation of this approach, both artificial and real cases demonstrated its effectiveness.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available