What Does Over-fitting A Model Mean ?

Over-fitting

Overfitting a model happens when the model becomes too specific and complex that it ‘fits’ the data on which it is ‘trained’ on (read training data).

But when applied on test data, it does not give as good a prediction accuracy. This generally occurs when the model has too many parameters that it models the noise.

How to avoid over-fitting ?

To avoid over-fitting, a model has to be simple and generic enough that is explains the ‘mean’ part of the training data so that the parameter estimates will hold on the test data too.

  1. When there is a choice of two models giving the same accuracy, pick the simpler one with fewer co-variates / parameters.
  2. Remove redundant features
  3. Try to use large data for training. When you dont have one, consider bootstrapping.

 

 

If you like us, please tell your friends.Share on LinkedInShare on Google+Share on RedditTweet about this on TwitterShare on Facebook