Efficient Feature Scaling for Support Vector Machines with a Quadratic Kernel

摘要

Choosing multiple hyperparameters for support vector machines has gained wide attention from researchers and this also provides a strategy for automatically selecting scaling factors of features. This paper proposes an efficient feature scaling method for support vector machines with a quadratic kernel. The proposed method alternately performs the standard SVM algorithm and the eigen-decomposition until some criteria are met. It is interesting to note that scaling factors of features can be analytically obtained for fixed support vectors. The experiments on a toy example, UCI data sets, and face images are carried out to demonstrate the feasibility and effectiveness of the proposed method.