While training a model, we should ensure that the model  does not suffer from

  1. Over-fitting    – high variance.
  2. Under-fitting – high bias.For example

screen-shot-2016-11-26-at-6-38-21-pm

Handling How to ?

  1. High Bias  : underfitting – training / test error – high => Add more features better result. [1]
  2. High variance : overfitting – train error -low /  test error – high
    => Feature reduction
    It’s possible that we have too many features and that the model is overfitting, reducing the features will make the model more flexible.
    => Add more training examples can also helpscreen-shot-2016-11-26-at-6-38-57-pm
    Figure. Handling under / over fitting

Ref:

  1. http://www.kdnuggets.com/2016/12/4-reasons-machine-learning-model-wrong.html
  2. http://scott.fortmann-roe.com/docs/BiasVariance.html

 

Advertisements