Resampling (crossvalidation + bootstrap)

Training error + test error
Usually training error is smaller
Bias – deviation from truth
Variance – deviation from average
Other methods (adjust the math) – Cp stat, AIC , BIC
Fit model on training set, then Fitted model then used to predict data from valudation set

Disadvantages:
Overestimate error
Test errors vary

[youtube https://www.youtube.com/watch?v=nZAM5OXrktY]