L5 - Regression - Model Complexity Flashcards
What is the relationship between model complexity (polynomial degree) and the error?
As the polynomial degree increases, the error decreases.
What happens if the regression model complexity is too high?
The model becomes overfit. This means the model learns the data as opposed to the underlying patterns of the data.
Thus the model will perform poorly on unseen data.
What is cross-validation?
A technique to detect model overfitting.
Trains and tests the model on subsets of the data, and provides an error value of the model.
How do we know if our model is at ideal complexity?
When the cross-validation error is at its lowest.
If your regression equation has too high polynomial degree, what can this lead to? What about too low?
Too high - Overfitting
Too low - Underfitting
When running K-fold cross validation, what does it tell us when the C-V error starts to increase again?
That we have found the optimal model complexity.
