Support Vector Data Description Based on Rescaled Hinge Loss
Support Vector Data Description(SVDD)is highly susceptible to outliers,resulting in poor robustness.To this end,a new support vector data description method(RH-SVDD)is proposed using the rescaled hinge loss function.First,the bounded rescaled hinge loss is used as slack variables to construct a hypersphere model;secondly,the hypersphere model is transformed into a convex optimization problem through the conjugate function theory;finally,the convex optimization problem is solved iteratively using half-quadratic optimization technology,and the weights are updated during the iteration process,thereby weakening the impact of outliers and improving robustness.Experimental results show that the proposed RH-SVDD has better performance advantages in classification tasks.
support vector data descriptionrescaled hinge losshalf-quadratic optimizationrobustness
万方数据
维普
