کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
406113 | 678060 | 2015 | 11 صفحه PDF | دانلود رایگان |
Robustness is an important characteristic of a classifier. With higher robustness, a classifier can resist much more against the noise of a contaminated dataset which is usually occurred in the real-world applications. With its excellence in robustness, a fuzzy support vector machine (fuzzy SVM) developed by Lin and Wang deserves the most attention among varieties of support vector machines. The main goal of this paper is to gain the Joachims’ ξ–α bound of the fuzzy SVM. Based on the decoupled α and ξ terms in its expression, the ξ–α based estimation is particularly suitable for robustness analysis of the fuzzy SVM. The study re-examines the theory of the fuzzy SVM having an additional fuzzy input si in details with the ξ–α estimation, and conducts a relatively contracted condition for upper bounding the corresponding performance. The bound confirms the crucial robustness which the fuzzy SVM can achieve analytically, and would be helpful for the works such as model selection or model adaption for further applications.
Journal: Neurocomputing - Volume 162, 25 August 2015, Pages 256–266