How to Choose Kernel in SVM
There is no one rule to define the kernel.
Domain experts can help to learn about the data distribution, which can be a factor to choose the kernel.
Choosing kernel for SVM depends on,
- Data distribution
- Size
- Data characteristics
- Noise
- Outliers
- If Linear, use standard Support Vector Machine (SVM)
- If Non-Linear,
- Use Cross Validation to find which one to use - Polynomial Kernel, Radial Basis Kernel