MDL Returns Rissanen's Minimum Description Length.
m=model that has been estimated using System Identification toolbox.
This function requires System Identification toolbox to work. Plug-compatible with built-in functions aic(m) and fpe(m).
MDL can be used like AIC or FPE to compare models of different complexities. Choose model with lowest MDL or AIC or FPE. Pintelon & Schoukens (2001) pp. 329,550 say MDL is better than AIC because AIC tends to select a too-complex model.
Minimum Description Length = V*(1+d*log(N)/N)
where V=loss function, d=number of fitted model parameters, and N=number of data points.