The weight of the bias component in the learning algorithm's error. I.e., the percentage of errors that can be attributed to bias error (underfitting) as opposed to variance error (overfitting).