2

I have two questions about isolation forest. I may not understand how it works correctly but I just wonder:

  1. What is the cost function of the isolation forest?
  2. What is the optimization method to minimize such a loss function?

I read the original paper and was looking for an answer in other sources but I could not find any answer. I just found that decision tree uses a MSE as a cost function while isolation free is based on randomization.

Sycorax
  • 90,934
Amhs_11
  • 333

1 Answers1

1

Same as with regular decision tree, isolation forest is not trained by directly minimizing some loss, but by using a dedicated algorithm. If you are interested in the algorithm, check the paper by it's authors, where they describe it in detail:

Liu, Fei Tony, Ting, Kai Ming and Zhou, Zhi-Hua. (2008). “Isolation forest.” Data Mining. ICDM’08.

Tim
  • 138,066