1

I'm studying machine learning and one of the key points I make sure is to understand the origin and the reasoning why such a name or term is used. It could be a rich and insightful dive into the concept it is trying to convey.

So, about bias-variance trade-off. Why use the word 'bias'?

In the usual, everyday parlance, 'bias' means leaning towards one side. If in the case of explaining pros and cons of a certain ideology, to be 'biased' means to give an unbalanced view and lean towards something.

Can we say the same for machine learning?

Thanks

Glen_b
  • 257,508
  • 32
  • 553
  • 939
cgo
  • 7,445
  • 10
  • 42
  • 61
  • 3
    BTW, this is not what is meant by a 'self-study' question. That refers mostly to homework or routine exercises. – mkt Sep 30 '19 at 09:01
  • 1
    Wikipedia adequately defines the *meaning* of (estimation) [bias](https://en.wikipedia.org/wiki/Bias_of_an_estimator) (& it *does* carry a sense of leaning toward one side); it has a section on bias, variance and MSE, and if you look at [Bias-variance tradeoff](https://en.wikipedia.org/wiki/Bias%E2%80%93variance_tradeoff) you'll see the direct connection to that section. As for its origins, the term has been around in statistical writing since the late 19th century: See [Earliest Known Uses of Some of the Words of Mathematics (B)](http://jeff560.tripod.com/b.html) (and then search on 'bias'). – Glen_b Sep 30 '19 at 10:01
  • ... and almost any reasonable text with a section on estimation will have extensive discussion of bias (among numerous other relevant properties of estimators); it's standard fare in statistics-degree programs. – Glen_b Sep 30 '19 at 10:10
  • You may also be interested in the electronics variety of the term bias: https://en.wikipedia.org/wiki/Biasing. – cbeleites unhappy with SX Sep 30 '19 at 13:53

0 Answers0