A practical approach to model selection for support vector machines with a Gaussian kernel.

Area of Science:

• Machine Learning
• Computational Statistics
• Pattern Recognition

Background:

• Support Vector Machines (SVMs) are powerful classification tools.
• Generalization ability of SVMs depends on learning parameters like kernel parameter (γ) and cost parameter (C).
• Model selection is crucial for optimizing these parameters.

Purpose of the Study:

• Propose a novel, computationally efficient model selection method for SVMs with a Gaussian kernel.
• Determine optimal values for the kernel parameter (γ) and cost parameter (C).
• Minimize central processing unit (CPU) time during model selection.

Main Methods:

• Develop a method to determine the kernel parameter (γ) based on feature space dimensionality and class dispersion.
• Utilize an analytical function for efficient kernel parameter retrieval.
• Evaluate the method on 17 diverse classification problems.

Main Results:

• The proposed method successfully identifies suitable kernel (γ) and cost (C) parameters.
• Achieves classification performance comparable to existing methods.
• Significantly reduces the computational effort required for parameter tuning.

Conclusions:

• The novel model selection method offers an efficient approach for SVMs with Gaussian kernels.
• It provides a favorable balance between computational cost and classification accuracy.
• The method's analytical approach simplifies parameter determination.