L2-norm multiple kernel learning and its application to biomedical data fusion

BMC Bioinformatics

Table 11 Comparison of the performance obtained by joint estimation of λ and standard cross-validation in LSSVM MKL

Data Set	Norm	Validation Approach	Estimation Approach
endometrial disease	L _∞	0.2625 ± 0.0146	0.2678 ± 0.0130
	L ₂	0.2584 ± 0.0188	0.2456 ± 0.0124
miscarriage	L _∞	0.1873 ± 0.0100	0.2319 ± 0.0015
	L ₂	0.1912 ± 0.0089	0.2002 ± 0.0049
pregnancy	L _∞	0.1321 ± 0.0243	0.1651 ± 0.0173
	L ₂	0.1299 ± 0.0172	0.1165 ± 0.0100

Comparison of the performance obtained by joint estimation of λ and standard cross-validation using LSSVM MKL. As shown, the estimation approach based on L₂ MKL is better than L_∞ MKL. This is because when the kernel coefficients are sparse, the estimated regularization parameters λ are either very big or very small, which are usually not optimal values in LSSVM. In contrast, the λ values estimated by L₂ method are at normal scale and often close to the optimal values.

ISSN: 1471-2105