1

I have a binary labelled dataset with numeric features. I want to create a "business rule" of the type y = x1 > t1 and x2 > t2 and x3 > t3. X are my features and I just have to find which are the thresholds t that fit the data better.

I have tried hyperopt to find the optimal set of 3 thresholds and it kindof works but it feels lazy and inefficient, I am interested in something smarter.

I have thought decision trees but I haven't managed to tweak the hyperparameters to build a tree with the shape of the rule I want.

Any ideas?

hipoglucido
  • 1,160
  • 1
  • 10
  • 17

1 Answers1

0

Decision trees will work to create a series of rules to predict a target value.

You don't get to pick the shape of the rules, aka the parameters. The decision tree divides into most pure regions based on the data. You can choose the hyperparameters (e.g., the depth of the tree).

Brian Spiering
  • 20,142
  • 2
  • 25
  • 102
  • There many ways to influence the shape of a decision tree beyond its maximum depth. I just don't think they give me enough control to do what I want. – hipoglucido Dec 24 '22 at 12:36