Simulation result of this type: Should I use a diferent flavor of logistic function?

Screenshot attached:
My Inference from the simulation output: No clear demarcation between benign and malignant. How to attack this problem now?
Screenshot from 2022-09-20 14-45-45
ched.

Hello @tennis_geek,

Considering those data points are what you have, it appears that the feature selected for the model isn’t good enough - that both benign and malignant tumor mostly share the same range of feature value. In this case i would suggest you to find a different feature that is more discriminative.

Raymond

Thanks @rmwkwok, options I could think of:

  1. As you suggested: [quote=“rmwkwok, post:2, topic:207336”]
    In this case i would suggest you to find a different feature that is more discriminative.
    [/quote]
  2. Try a higher order polynomial in defining logistic function.
  3. both the above

For number 2, if a benign sample and a malignant sample both have a tumor size s of 3, then even if we create a higher order version of the tumor size, say s^3, the two said samples will still share the same value for that new, higher order feature which is 3^3 = 27. The new feature is still not more discriminative, is it?

Raymond

If the turning point of the polynomial function can be adjusted less steeply to accomodate more demarcation, might be a better option. Still not a robust option as compared to a domain-specific discriminative feature which might solve the problem better. But can give it a try using this combo in my opinion. Especially in scenarios where no more reliable distinctive features are present to include in the model.

I get the idea.

Raymond