Akaike Information Criterion

• Considers goodness-of-fit to the data and penalizes complexity of the model
• $AIC=-2log(L)+2q$
• where:
• L: likelihood function for a particular model
• q: number of variables of this model
• If error terms $ϵ$ follows Normal Distribution , expected value 0 + constant variance $AIC=\frac{1}{\eta {\sigma }^2}(RSS+2p{\hat{\sigma }}^2)$