Analyze model fairness with Fairlearn

Fairlearn is a Python package that you can use to analyze models and evaluate the disparity between predictions and prediction performance for one or more sensitive features.

It works by calculating group metrics for the sensitive features you specify. The metrics themselves are based on standard scikit-learn model evaluation metrics, such as accuracy, precision, or recall for classification models.

The Fairlearn API is extensive, offering multiple ways to explore disparity in metrics across sensitive feature groupings. For a binary classification model, you might start by comparing the selection rate (the number of positive predictions for each group) by using the selection_rate function. This function returns the overall selection rate for the test dataset. You can also use standard sklearn.metrics functions (such as  accuracy_score, precision_score, or recall_score) to get an overall view of how the model performs.

Then, you can define one or more sensitive features in your dataset with which you want to group subsets of the population and compare selection rate and predictive performance. Fairlearn includes a MetricFrame function that enables you to create a dataframe of multiple metrics by the group.

For example, in a binary classification model for loan repayment prediction, where the sensitive feature Age consists of two possible categorical values (25-and-under and over-25), a MetricFrame for these groups might be similar to the following table:

Visualizing metrics in a dashboard

It's often easier to compare metrics visually, so Fairlearn provides an interactive dashboard widget that you can use in a notebook to display group metrics for a model. The widget enables you to choose a sensitive feature and performance metric to compare, and then calculates and visualizes the metrics and disparity, like this:

Integration with Azure Machine Learning

Fairlearn integrates with Azure Machine Learning by enabling you to run an experiment in which the dashboard metrics are uploaded to your Azure Machine Learning workspace. This enables you to share the dashboard in Azure Machine Learning studio so that your data science team can track and compare disparity metrics for models registered in the workspace.

Sources

Microsoft Learn

#modelling #datascience #machinelearning

Article By Devansh Mistry

LinkedIn                           GitHub                               Devansh Mistry