Under and over fitting

A clear non-linear relationship applies between x and y, fitting a linear trend (polynomial of degree 1) does not capture the observed variation, fitting a very high order polynomial drives the fitted function through individual observations but leads to extreme predictions outside these observations.