1
$\begingroup$

I have just started learning about machine learning, and I started reading about handwriting recognition. Apparently (please correct me if I am utterly wrong), handwriting is treated as a classification problem. It makes sense because you are assigning a certain value depending on other factors (e.g. two vertical lines + one horizontal line = "H"). Now, if we were to treat it as a regression problem, would it not work just as fine? It would try to predict the value of the letter by looking at the patterns - just like classification.

Please, correct me if I am wrong. I am just really confused right now.

$\endgroup$
  • 2
    $\begingroup$ This may seem a bit rude, but could you include in your question what the words "classification" and "regression" mean to you? $\endgroup$ – Matthew Drury Aug 25 '17 at 20:15
  • 2
    $\begingroup$ @Aksakal I know what it means : ) I just want us to know the OP's understanding, so we can better help out. $\endgroup$ – Matthew Drury Aug 25 '17 at 20:29
  • $\begingroup$ Imagine yourself trying to work out what handwriting you're having trouble reading says (like a hastily scrawled name and address, say) ... answering a question like "Is that an '$a$' or a '$c$' or an '$o$'?" is classification. $\endgroup$ – Glen_b Aug 26 '17 at 0:54
  • $\begingroup$ @Glen_b Thanks for that concise explanation! I was a little bit confused about the terms. $\endgroup$ – Armando H. Aug 30 '17 at 17:39
  • $\begingroup$ @MatthewDrury I had understood that classification was used to get a value of either 1 or 0 (is it a tumor or is it not?), but I did not know that more terms could be used when I asked the question (What type of tumor?). As I understand, regression is used to predict future values of certain elements based on the values given in a data set. $\endgroup$ – Armando H. Aug 30 '17 at 17:43
4
$\begingroup$

The machine learning textbook explanation would be that your're mapping the inputs to a discrete set of letters, that's why it's classification. The outcome variable is not only discrete, but also nominal or categorical. In that 'a' is not more or less than 'b' or 'c', it's simply distinct, different from other values.

If you were to map the writing into continuous variable, then they'd say it's regression. For instance, if you were trying to assess the stress level by hand writing, where stress level is a continuous score from 0 to 1. Here, not only the outcome is continuous, it's also cardinal, in that 0 is actually less than 1, not just different. Moreover, 0 is at the same distance from 0.5, as 0.5 from 1. In classification problems this is not the case, you simply deal with distinct outcomes, they often not even ordered

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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