Machine learning for public policy making : how to use data-driven predictive modeling for the social good

Steuer, Fabian

Machine learning for public policy making : how to use data-driven predictive modeling for the social good

License

Citation

Steuer, F. Machine learning for public policy making : how to use data-driven predictive modeling for the social good. 2018. handle: http://hdl.handle.net/10230/43488

This citation was generated automatically.

Abstract

Machine learning gives computers the ability to learn from data without being explicitly programmed. Due to its excellent prediction abilities, it has recently gained traction in economics, statistics and social sciences. Real-world problems machine learning has been applied to include predicting the probability that individuals commit crimes, targeting hygiene inspections by data-mining online restaurant reviews or estimating poverty levels based on satellite imagery. In this thesis I explore how machine learning can help to solve such and other prediction problems in public policy making and what challenges it faces. My goal is to bring the two fields closer together as most public policy makers likely do not even know that they face prediction problems that machine learning can help solving. After an introduction to prediction problems, I give an overview of how machine learning works and explain under what circumstances machine learning can be used for data-driven predictive modeling for the social good. A case study about predicting hygiene violations in restaurants illustrates the lessons learned and allows to get an idea of what applying machine learning looks like in practice. I then look into the challenges and limitations that machine predictions face in public policy making. Besides the fundamental limits of prediction, these range from technical and human challenges to ethical and legal issues due to biased predictions, black-box algorithms and questions of responsibility.