Caesar AI Atlas

Prediction Bias

Caesar AI Atlas Definition

Prediction bias is the difference between the average of a model’s predictions and the average of the corresponding labels. It is a diagnostic signal about systematic overprediction or underprediction and is distinct from ethical bias or a model’s bias parameter.

Other Definitions

Prediction Bias Source

A value indicating how far apart the average of predictions is from the average of labels in the dataset. Not to be confused with the bias term in machine learning models or with bias in ethics and fairness.

Related Terms