|
Hello, I have a problem which requires the use of an SVM. However, there is a feature (in 100 features) that takes the all range of values (allowed for that feature) in each class. for example (imaging this feature as the 4th feature, and x taking any random value) - Class [feature vector] - 1 1:x 2:x 3:x 4:0.0 - 1 1:x 2:x 3:x 4:... - 1 1:x 2:x 3:x 4:1.0 - 2 1:x 2:x 3:x 4:0.0 - 2 1:x 2:x 3:x 4:... - 2 1:x 2:x 3:x 4:1.0 - 3 1:x 2:x 3:x 4:0.0 - 3 1:x 2:x 3:x 4:... - 3 1:x 2:x 3:x 4:1.0 This feature is important because the all other n features changes depending on the value of this feature. This means that probably will exist two classes with the same feature vector and the only difference will be this feature. From the mathematical perspective and from the SVM illustration, there will exist an hipper-plane that splits the two bases based only in this features. However from a practical implementation, what should be the right approach to deal with this feature?
- give to this feature more weight?
- Split this feature into n clutters and create an SVM for which cluster? (do not use this input as a feature, but use as an SVM selector ) Thanks,
Filipe |
