Explainable Machine Learning for Public Policy
Machine learning may provide opportunities for governments to make better decisions. However, with its inscrutability, it raises new challenges for accountability, which is essential for public policy. We built and iteratively improved a web application for analyzing logistic regression models. We ran focus groups to understand how the application would be used in public policy discussions. We found many ways to improve machine learning visualizations, but also found that an application that both lets users understand how a model works must also explain the concepts of the model. Our application was useful for public policy discussions among users experienced in machine learning, and with added educational tools, the application could be used by a wider audience.