- AIC (Akaike Information Criterion)
- BIC (Bayesian Information Criterion)
Equation
$l:\text{likehood, (means how strong the model is to fitting the data)}$ - 愈大愈好
$k:\text{number of parameters}$ - 愈小愈好
$n:\text{number of samples used for fitting}$ - 愈小愈好
$AIC=2k-2l$
$BIC=ln(n) \cdot k -2l$
- Lower AIC means higher log likehood (l) or less parameters (k)
- Lower BIC means higher log likehood (l) or less parameters (k) or less samples used in fitting
Code Reference
https://github.com/ritvikmath/Time-Series-Analysis/blob/master/Model Selection.ipynb
Video Reference
https://www.youtube.com/watch?v=McEN54l3EPU