 Why is the mean squared error so good and a lot of machine learning we tend to minimize the mean squared error and Well, the mean squared error is so good because it considers both a bias component and a variance component So it will only go down very low if both the bias and the variance are also low a Low bias indicates that the parameters learned by the model are actually correct on Average and the variance indicates how much the parameter values Change with changes in data set across different sets of training We want both of these to be low in order to have good models This irreducible error is due to inherent noise in the nature of real data Minimizing MSE may not be able to reduce this error But by reducing bias and variance gives good model results for generalizing data