Fast and efficient strategies for model selection of Gaussian support vector machine

Area of Science:

• Machine Learning
• Computer Vision
• Pattern Recognition

Background:

• Support Vector Machines (SVMs) are powerful classification algorithms.
• Effective parameter selection (kernel parameter sigma, penalty coefficient C) is crucial for SVM performance.
• Traditional methods like 5-cross-validation (5-CV) can be computationally intensive.

Purpose of the Study:

• To propose direct and heuristic strategies for selecting SVM kernel parameters and penalty coefficients.
• To enhance the efficiency and generalization capabilities of Gaussian SVMs.
• To offer a robust alternative to existing parameter selection methods.

Main Methods:

• A direct formula for kernel parameter (sigma) selection based on visual scale and data structure preservation.
• A heuristic algorithm for penalty coefficient (C) selection using classification extent in Sequential Minimal Optimization (SMO).
• Experimental evaluation on benchmark and real-world datasets comparing against 5-CV and radius-margin bound (RM) methods.

Main Results:

• The proposed strategies demonstrate superior efficiency and generalization compared to 5-CV and RM methods.
• Experiments on 13 benchmark and 3 real-world datasets validate the effectiveness of the new approaches.
• The new methods exhibit uniform and stable performance across diverse datasets.

Conclusions:

• The novel strategies for Gaussian SVM parameter selection offer significant advantages in efficiency and generalization.
• These methods provide a more effective and stable approach to optimizing SVM models.
• The findings suggest a promising direction for improving machine learning model performance through advanced parameter tuning.