(ML Spec)
When the model fails to capture the relationship between features and output, even on the training set. It’s said to have high bias It’s the opposite to overfitting
Approach
Make the learning algorithm more powerful/complex to deal with training data
- Try getting additional features
- Try adding polynomial features (e.g.
) - Try decreasing regularization parameter
(less penalty for model complexity) - Use a bigger artificial neural network with proper regularization
Note that these approaches usually come up with higher computational cost,