0
$\begingroup$

Suppose you have a supervised learning project where it is not easy to check whether the value you predicted is correct or not. So, in this case, does it still make sense to talk about the classification error? If not, what is the possible way to check whether the predicted value is correct or not? By comparing the predicted results with the corresponding experimental results? Except this, are there other possible ways to do it?

$\endgroup$
  • 1
    $\begingroup$ If you do not have any data with correct labels/outputs, then this is not a supervised problem. It is not clear, what you are asking. You may want to provide an example. $\endgroup$ – lanenok May 27 '15 at 2:54
  • $\begingroup$ @lanenok, thanks. I do have some data with the correct labels/outputs. The model I cerate will definitely satisfy this training set without any doubt. The thing is when I use the model to predict, I do not know whether the model would predict correctly or not. $\endgroup$ – Joseph Stone May 27 '15 at 13:20
  • $\begingroup$ after reading the article with title " a few useful things to know about machine learning", I think maybe I need to set aside a few data with labels first, create the classifier, and then test the classifier on the set-aside data first to see the classification error. What do you think? $\endgroup$ – Joseph Stone May 27 '15 at 14:32
1
$\begingroup$

Randomly divide your training data into 10 subsets, hold out each one while training on the rest, test each classifier on the example it did not see and average the result in the end.

Reference: a few useful things to know about machine learning

$\endgroup$
  • 1
    $\begingroup$ AKA ten-fold cross validation :) $\endgroup$ – stochazesthai May 27 '15 at 15:12

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.