Model Bias

Definition:

The tendency of a model to shape or produce results that deviate from what would be the most likely observed outcomes in the real world when faced with the same conditions as the model.

This kind of tendency can exist whether intentional or not and whether it is due to the structure and assumptions of the model, or the process of selecting and preparing input data for the model.

Commentary:

The results of an analytics model are based on the data that it is presented with, either when being created and trained or when running in production. Bias arises when that data is not representative of the real world. The data may be missing key variables that would lead to different decisions, or it may include human-produced content that incorporates unrecognized biases of those persons.

Bias can also describe how well a model matches the training set. A model with high bias will not match the training data set closely, while a model with low bias will match that data set very closely.

Sometimes bias comes from models that are overly simple and fail to capture the trends present in the data set.