Akaike information criterion

Akaike information criterion, AIC

From Wikipedia, the free encyclopedia

The Akaike information criterion is a measure of the relative goodness of fit of a statistical model. 

It was developed by Hirotsugu Akaike, under the name of "an information criterion" (AIC), and was first published by Akaike in 1974.^[1] 

It is grounded in the concept of information entropy, 

in effect offering a relative measure of theinformation lost when a given model is used to describe reality. 

It can be said to describe the trade off between bias and variance in model construction, 

or loosely speaking between accuracy and complexity of the model.

AIC values provide a means for model selection. 

AIC does not provide a test of a model in the sense of testing a null hypothesis

; i.e. AIC can tell nothing about how well a model fits the data in an absolute sense. 

If all the candidate models fit poorly, AIC will not give any warning of that.

http://en.wikipedia.org/wiki/Akaike_information_criterion