Lasso Regression

A Simple Explanation - By Varsha Saini

If your machine learning model learns the pattern of data very well while training, it can lead to the problem of Overfitting. Overfitting is a situation in which models perform very well on the training data and poor on testing data. To resolve this issue, a penalty term can be added to the equation of the model, this process is called Regularization.

Lasso Regression

  • Lasso Regression is a regularization method in which a small bias is added to the linear equation such that the overall complexity of the model is reduced.
  • It stands for Least Absolute and Selection Operator.
  • It is also called L1 Regularization since the penalty added is absolute value(power 1).
  • It is used as a method for Feature Selection, the features for which the weight is very low can be dropped.
  • It can handle data that suffers from Multicollinearity problem.

Cost Function of Linear Regression

  • Cost =

Cost Function of Lasso Regression

The cost function of Linear Regression is modified by adding a Regularization term to it.

  • Cost =
  • slope m is the coefficients (weight) of the independent features.
  • lambda  λ is the penalty term.


  • Regularisation parameter is added to the cost function such that when an optimization algorithm like gradient descent is used to minimize the cost function, the values of weights are reduced.
  • Reducing weights decreases the complexity of the model, reduces variance, and hence resolves the Overfitting problem.

Controlling the Regularization Parameter

  • Lambda λ which denotes the penalty term in the equation can be used to control the regularization parameter.
  • A high value of lambda increases the regularization parameter and decreases the value of coefficient or weight. It can increase bias and lead to the problem of Underfitting.
  • A low value of lambda reduces the regularization parameter and does not have much effect on the value of coefficient or weight. Hence the problem of Overfitting may not get resolved as variance will not be reduced.
  • Finding an optimal value of λ is a must to achieve a perfect Lasso Regression model.