Bias detection and mitigation in ML Python - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is bias in machine learning?

Bias in machine learning means the model unfairly favors or disfavors certain groups or outcomes, often due to unbalanced or incomplete data.

beginner

Name one common cause of bias in machine learning models.

One common cause is biased training data, where the data does not fairly represent all groups or scenarios.

beginner

What is bias mitigation?

Bias mitigation means using methods to reduce or remove unfair bias from a machine learning model to make it more fair.

intermediate

Give an example of a bias detection technique.

One example is measuring model performance separately for different groups to see if accuracy or errors differ significantly.

intermediate

What is one way to mitigate bias during model training?

One way is to balance the training data by adding more examples from underrepresented groups.

What does bias in a machine learning model usually indicate?

A The model treats some groups unfairly

B The model has perfect accuracy

C The model runs faster

D The model uses more data

Bias means unfair treatment or errors that affect some groups more than others.

Which of these is a common cause of bias?

A Balanced and diverse data

B Increasing model layers

C Using more computing power

D Biased or incomplete training data

Biased or incomplete data can cause the model to learn unfair patterns.

What is bias mitigation?

A Adding more bias to the model

B Reducing unfair bias in the model

C Making the model slower

D Ignoring bias in data

Bias mitigation means reducing unfair bias to improve fairness.

How can you detect bias in a model?

A Check if the model performs equally well for all groups

B Only look at overall accuracy

C Use less data

D Ignore group differences

Checking performance across groups helps find bias.

Which method helps mitigate bias during training?

A Removing underrepresented groups from data

B Using fewer features

C Balancing data with more examples from all groups

D Training for fewer epochs

Balancing data helps the model learn fairly from all groups.

Explain what bias in machine learning is and why it matters.

Think about how unfair treatment can happen in predictions.

Describe two ways to detect and two ways to mitigate bias in a model.

Consider both checking the model and changing data or training.