A fast leave-one-out cross-validation for SVM-like family. Issue 6 (August 2016)

• Record Type: Journal Article
• Authors: Zhang, Jingxiang; Wang, Shitong
• Is Part Of: Neural computing & applications. Volume 27:Issue 6(2016)
• Page Start: 1717
• Page End: 1730
• Publication Date: 2016-08
• Subjects: Leave-one-out (LOO) -- SVM-like family -- Adjustment factor -- Fast leave-one-out (FLOO); Neural networks (Computer science) -- Periodicals; Neural circuitry -- Periodicals; Artificial intelligence -- Periodicals; Neural Networks (Computer) -- Periodicals; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux nerveux -- Périodiques; Intelligence artificielle -- Périodiques; 006.32
• DOI: 10.1007/s00521-015-1970-4
• Languages: English
• ISSNs: 0941-0643

Abstract

Abstract The leave-one-out cross-validation is an important parameter selection strategy for SVM-like family, including SVM and SVR. However, due to the high computational complexity, the adaptability of this strategy is restricted. In this paper, aiming at its practical application, a fast leave-one-out cross-validation method by using an adjustment factor is proposed which focusses especially on the practicability for the SVM-like family where the decision function can be expressed by explicit dot-product of each training sample pair. The ability of the proposed method in fast parameter selection is better than that of original leave-one-out cross-validation with the same or comparable learning performance. The simulation results indicate the effectiveness and speedup of the proposed leave-one-out cross-validation method.