Underfitting in a Neural Network

  • Underfitting is opposite of overfitting
  • The model will not even be able to classify on the data on which it is trained
  • Training loss is very high
  • Training accuracy is very low
 

Techniques to reduce underfitting
  • 1 ) Increasing the complexity of our model 
  • like by increasing number of layers
  • like increasing number of neurons 
  • 2) Add more features to the dataset in our training set
  • 3) Reducing dropouts



Last modified: Monday, 11 January 2021, 12:21 AM