Assess the potential bias in a computational model

ID: B2AI_USECASE:43

Name: Assess the potential bias in a computational model.

Description: Even a high-performing computational model may be subject to bias, both explicitly and implicitly. The model may not accurately represent the population it was trained on. It may be algorithmically biased, with some features gaining weight over others in unexpected ways. There may be biases resulting from human preconceptions, e.g., human curators may have already made assumptions about disease status of patients contributing data to the model’s training set. There may also be unexpected confounders, such as social or economic factors contributing to clinical outcomes.

Category: assessment

Involved in: Quality Control

Data Topics:

• B2AI_TOPIC:5 (Data)

Contributor: Harry Caufield (ORCID:0000-0001-5705-7831)