In order to do linear regression you fit a linear formula to the data WX + b and in order to minimize the MSE you make the weighted cost function and try to reduce it with many ways, common one is gradient descent algorithm.