In classification tasks, it is easy to construct a confusion matrix, which shows many samples were classified correctly (true and false positives), and how many samples were classified incorrectly (true and false negatives). The various metrics that can be computed from the confusion matrix are quite easy to understand.

What if my target variable is continuous (e.g. if I am predicting the height of a person based on some genetic, environmental, etc. data)? What will be my "chance level accuracy"? How do I analyze the correctness of the classifier?

  • 3
    $\begingroup$ I would argue that it's much clearer to evaluate accuracy of a continuous variable. If the person's true height is 50cm, and 53cm was predicted. We know exactly how far off it is. We do not get that same effect with Yes/No responses. $\endgroup$ – Pierre L Sep 6 '16 at 5:57
  • $\begingroup$ It’s actually no longer a classifier, but a regressor. The rsme, as mentioned, is useful because it’s in the same unit as your original variable $\endgroup$ – user0 Jan 13 '18 at 16:49

In case of continuous variables, you can directly measure the difference from the actual value. Statistics like the Root Mean Square Error(RMSE) would then be useful to determine how far from reality your predictions are.

"Chance level accuracy" could be looked at as RMSE if in place of making predictions you simply used the mean.


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.