1
$\begingroup$

I would like to know why logistic function is chosen for classification.

After reading the original paper developing logistic regression by David Cox, I can understand the benefits of using the logistic function. However, in the 1950's, were there alternatives to the logistic function could help David Cox? If there was any, what would it be?

Improve this question
$\endgroup$
1
  • 3
    $\begingroup$ Not wanting to undermine Sir David's reputation, but I think he would want to underline the contribution of Berkson in 1944 and other work at the same time in similar style. See e.g. papers.tinbergen.nl/02119.pdf for some history. $\endgroup$
    – Nick Cox
    Commented Jan 30, 2018 at 8:10

1 Answer 1

Reset to default
4
$\begingroup$

It depends on what kind of model you're looking for, I suppose. Interestingly, a well-known discriminative classifier was also developed in the late 1950s - Rosenblatt published his paper on the Perceptron in 1958. However, it's formal properties (the Perceptron convergence theorem) would not be proven until 1962.

For a more classical statistically-flavored approach, Fisher's work on linear discriminant analysis was published well before Cox's work, in 1936.

Improve this answer
$\endgroup$
2
  • 1
    $\begingroup$ To add to this, Naive Bayes classifier was probably invented pretty early on as well. $\endgroup$
    – Francis
    Commented Jan 30, 2018 at 7:10
  • $\begingroup$ Oh, that's a good point! In theory the groundwork for the Naive Bayes classifier had been laid down as early as the 18th century with the work by Bayes himself, and as a classification technique it must have existed as folk knowledge from sometime in the early 20th century. Interestingly, it seems that the precise first usage in a supervised learning-like setting is hard to pin down: stats.stackexchange.com/questions/18212/… $\endgroup$
    – Louis Cialdella
    Commented Jan 30, 2018 at 8:27

Your Answer

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

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