Akaike Information CriterionAkaike's Information Criterion
(Japanese)The Akaike Information Criterion (Japanese: Akaike jōhōryō kijun; originally called An Information Criterion and later Akaike's Information Criterion) is an index used to evaluate the quality of statistical models. It is commonly abbreviated as AIC, which is the more widely used name.
It is a very well-known index in statistics and is implemented in many statistical software packages. It was devised in 1971 by Hirotugu Akaike, former Director-General of the Institute of Statistical Mathematics, and presented in 1973.
AIC is used to balance model complexity against goodness of fit to the data. For example, consider building a model to explain measurement data statistically. In general, the more parameters or model terms included, the better the fit to the observed data.
However, adding too many terms can force the model to fit incidental variation such as noise that is unrelated to the underlying structure of the phenomenon, causing the model to perform poorly on other data of the same type (the problem of overfitting). To avoid this, the number of modeled parameters should be kept as small as reasonably possible, although determining the appropriate number is not easy. In practice, selecting the model with the smallest AIC often leads to a good model choice.
(English)
The Akaike Information Criterion (originally An Information Criterion, later called Akaike's Information Criterion) is an index for evaluating the quality of a statistical model. It is also called simply AIC, and this is the more common name. It is a very well-known indicator in the world of statistics and is provided in many statistical software packages. It was invented by Hirotsugu Akaike, former director of the Institute of Statistical Mathematics, in 1971 and published in 1973.
The AIC is used to "balance the complexity of the model with the goodness of fit to the data. For example, consider creating a model that statistically explains some measurement data. In this case, the greater the number of parameters and order, the better the goodness of fit to the measurement data. However, on the other hand, the model will not fit the same type of data because it will be forced to fit accidental (unrelated to the structure of the measurement object) variations such as noise (overfitting problem). To avoid this problem, the number of modeling parameters must be kept to a minimum, but the actual number is a difficult question. Specifically, selecting the model with the smallest AIC will often result in the selection of a good model.
Contact Us
Over 30,000 consultations / Over 19,000 completed engagements. To date, we have supported consultations and requests involving analysis outsourcing, statistical processing, questionnaire surveys, marketing support, and more. Our experienced consultants carefully listen to your needs so that we can provide the right support. We offer prompt and accurate work at reasonable, accessible rates and are committed to delivering dependable results. Stat Agent team members across Japan will take responsibility for supporting you.
*We provide the profile of the person responsible for your project when you apply.
For outsourced analysis and statistical processing, choose Stat Agent.
0476-85-7930
*When order volume is high, it may be difficult to reach us by phone.
We apologize for the inconvenience. We respond in order of receipt, so if your matter is urgent, please contact us byemail .
*When order volume is high, it may be difficult to reach us by phone.
We apologize for the inconvenience. We respond in order of receipt, so if your matter is urgent, please contact us by
