Knowledge MaxxingUnderfitting
Underfitting happens when a model is too simple to capture the real patterns in the data, performing poorly on both training and new data.
Prerequisite: Overfitting
What matters
Unlike overfitting, underfitting shows up as weak performance everywhere, including on the training set itself, which is the clue that points to the model needing more capacity or better features.
Example
Fitting a straight line to data that clearly curves will underfit, since a straight line cannot represent the curved relationship no matter how it is trained.
Common misconception
Underfitting is not fixed by gathering more training data alone; if the model itself is too simple, more data of the same kind will not help much.
Recall question
How does underfitting's symptom differ from overfitting's on the training set specifically?