A multi-label learning based kernel automatic recommendation method for support vector machine.

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• Kernel selection is critical for Support Vector Machine (SVM) classification performance.
• Existing methods often focus on kernel construction or parameter tuning, neglecting efficient kernel selection.
• Current cross-validation approaches for kernel selection are time-consuming and overlook computational costs like CPU time.

Purpose of the Study:

• To develop an automated kernel recommendation method for SVM classification.
• To address the tradeoff between classification accuracy and computational efficiency (CPU time).
• To propose a multi-label learning approach based on data characteristics for kernel selection.

Main Methods:

• A meta-knowledge database is created by extracting data characteristics and identifying applicable kernels.
• A multi-label classification model is trained on this database.
• The model recommends appropriate kernels for new datasets based on their characteristics.

Main Results:

• Extensive experiments were conducted on 132 UCI benchmark datasets.
• The proposed method demonstrated superior performance compared to existing kernel selection techniques.
• Recommended kernels by the proposed method achieved higher classification accuracy than the widely used Radial Basis Function (RBF) kernel.

Conclusions:

• The multi-label learning-based kernel recommendation method effectively identifies optimal SVM kernels.
• The approach balances classification accuracy with computational efficiency.
• This method offers a significant improvement for SVM classification tasks by automating kernel selection.