• What are the meanings of Bias?

  • And is Under fitting, which is used in machine learning contexts, the same as "Bias"?

I have faced biased data in sampling in statistics but it seems this is a different thing to bias in learning concepts.

I have heard that some data sets are biased, also have heard the model (for example neural network) has low bias or e.g. 'high bias' problem. Are these uses of bias different?


2 Answers 2


Bias can mean different things in statistics:

  • If your model is biased, it's likely your model is under-fitting.
  • Some data set is biased in sample collection. For instance, if you assume your sample responses are independent, but somehow it's not, this is a bias in your data set. If you want to sample everybody in the country, but you skip some cities for no reason, this causes bias in your data set.
  • Your estimators could be biased - the expectation of your estimator is not equal to the true value in the population.
  • "Bias" is also used to describe an learnable offset parameter when using transfer functions, e.g. in neural networks when calculating activation of an artificial neuron.
  • 1
    $\begingroup$ Bias is under fitting and variance is overfitting --- please review your answer. Unbiased is not a problem $\endgroup$ Commented Apr 5, 2017 at 6:10
  • $\begingroup$ @NeilSlater Ok. No need to give -1. I messed up the orders. EDITED. $\endgroup$
    – SmallChess
    Commented Apr 5, 2017 at 6:13
  • $\begingroup$ @SmallChess thanks for your answer, I searched alot but did not find valuable stuff. $\endgroup$ Commented Apr 5, 2017 at 7:58
  • $\begingroup$ Please read your answer next time before submitting. Thanks for editing. $\endgroup$ Commented Apr 5, 2017 at 8:01
  • $\begingroup$ Feel free to edit the answer. $\endgroup$
    – SmallChess
    Commented Oct 7, 2017 at 17:46

Bias is mainly how much error is there which can be caused due to empty fields or Nan values and under fitting will usually be caused if our training data will not have too much information which can help to train our model properly . A sample is biased if individuals or groups from the data are not represented in the sample as mentioned above. These are all related with statistics.


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.