Search code examples
wekadecision-treej48c4.5

WEKA J48 decision tree with non linearly separable data

Does Weka J48 Decision Tree classifier support classification for a problem with intrinsically non linearly separable data? In short, is J48 either a linear or a non linear classifier?

Solution

  • In short, is J48 either a linear or a non linear classifier?

    • I don't know.

    However, the decision boundaries of J48 can be made, in a way, "stepwise linear". So you can approximate a nonlinear decision boundary if you set minNumObjects low enough and set pruning to false (= unpruned to true).

    Here is a visualization of the linear boundaries using weka's BoundaryVisualizer:

    enter image description here

    This is accessible via the small rectangular "Weka GUI Chooser" window. There you can load your own dataset.

    Here is a video that explains it with more details: https://youtu.be/fMxfUKYkcVg?t=568