Discriminant Analysis , Logistic Regression Analysis and Support Vector Machine use a line (super plane) to classify samples.
But there are cases that a simple line is not valid to classify. The right figure is an example that both "a line" and "simple line" are not valid.
Kernel method is a method to use a simple line for such case.
If the areas are clearly separated, "changed data" may be good to use a simple line to classify. Kernel method uses this idea.
Kernel method is a kind of Analysis Using Intermediate Layer.
The image of kernel method is Analysis Using Intermediate Layer. But we do not need to calculate intermediate variables. It is called "Kernel trick",
We do not need to select the method to change data. But we need to select kernel function.
In the calculation of multi-variable analysis, there is a part of inner product. Kernel method uses kernel function as the inner product.
Kernel method is difficult for the use of Statistical Way of Making Hypothesis because it does not tell the factor of the classification.
Kernel trick is not good for all uses of analysis.
We can use kernel method by RapidMiner. It is easy because we do not need programming.
Difference of Output by Methods
NEXT One-Class SVMTweet