I am trying to understand linear regression using maximum likelihood approach. According to Bishop’s Patterns Recognition, a target variable y can be modeled by introducing an error term e that is normally distributed with mean 0 and some variance. Then the relation is given as y=h(x,w)+ e.