I'm trying to do this classification problem, depicted in the following Figure. The task is to separate the blue elements from the red elements in a cartesian (x,y) coordinate system. I have to:

● Find a transformation to map the 2D points in 3D points (by adding one component) that is able to make the data linearly separable.

● Make a graph (by hand) to show the transformed 3D space.

● Assuming that a Support Vector Machine (SVM) classifier is used to classify those points, add the separating hyperplane in the depicted plot.

● Explain how SVMs work. What are kernel functions?

● Explain the use of the C parameter in building a SVM model. What happens if C=0+ (a small positive value)? What happens if C=Infinite?


  • 3
    $\begingroup$ This looks like an assignment problem. What have you tried, where did you get stuck? $\endgroup$ – TYZ Mar 24 at 14:34
  • $\begingroup$ I have no idea which kernel and transformation i have to use to solve the problem $\endgroup$ – steph Mar 24 at 14:42
  • $\begingroup$ Can you give me an advice? $\endgroup$ – steph Mar 24 at 14:47
  • $\begingroup$ Someone have an idea or a tip? $\endgroup$ – steph Mar 24 at 15:22
  • $\begingroup$ You have studied SVM right? is there something you don't understand about it? $\endgroup$ – Erwan Mar 24 at 17:26

@TYZ, @Erwan https://www.kdnuggets.com/2017/08/support-vector-machines-learning-svms-examples.html/2 Maybe here there's something useful, but I don't know how one could derive the transformation sqrt2 x1x2 without reading it somewhere

| improve this answer | |
  • $\begingroup$ Multiplying x1 and x2 to get x3 isn't a terribly complex transformation, and it's certainly not the only one that would work - absolute values are also handy here. Heck, you could define a piecewise function defined only over those 16 points which would work. $\endgroup$ – Nuclear Wang Mar 24 at 18:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.