I have tabular dataset each row represent the curve, so goal is to filter out curves that do not follow Sigmoid function. Obviously first I can label the curves.

My question would be is machine learning (classification , LSTM ) applicable here or better than the conventional coding solution ?

  • $\begingroup$ Parameterize them in some way. What if you try and fit a sigmoid (logistic) to each curve and use the parameters as features? There are many examples of this and python scipy will fit the curve for you for arbitrary non-linear functions. I would then look to cluster or try some kind of anomaly detection. $\endgroup$ Nov 4, 2021 at 19:46
  • $\begingroup$ Each curve has 42 data points, may be I can use each of them as feature ? $\endgroup$
    – alex3465
    Nov 5, 2021 at 9:17
  • $\begingroup$ But isn't this fine? When it's not a sigmoid, you will get parameters that are different. Have you tried plotting the parameters that result in a 2D space for instance? I bet you get some nice clusters. $\endgroup$ Nov 5, 2021 at 13:19
  • $\begingroup$ @neuroguy123 parameters you mean popt or pcov $\endgroup$
    – alex3465
    Nov 5, 2021 at 17:24
  • $\begingroup$ Yes, what happens if you plot the 4 parameters? Do they distinguish what you want? $\endgroup$ Nov 5, 2021 at 17:32

1 Answer 1


One option is curve fitting. Fit a sigmoid function to the data and decide if the quality of fit is below a threshold for goodness of fit.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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