Achieving Fairness in Zoning Laws with Machine Learning

William Schimitsch, Farhad Mohsin

Keywords: Zoning, counterfactual analysis, fair assignment

TL;DR: In order to achieve fair usage and reassignment of land, we propose a computational approach and perform counterfactual analysis to find issues with current zoning practices.

Abstract: Zoning is a powerful regulatory tool used to determine the land use and development of a given area. Decisions regarding zone designations, then, have large implications for any given community. It is a problem where local zoning decisions are made by a small-sized zoning board, often through an opaque process. We note two aspects of the task of zoning; first, the end task is to assign a equitable assignment of zoning class (e.g., residential, commercial, mix-use etc.); and second, we have a large amount of data (geographical, demographic, and infrastructural) that is relevant to zoning. Thus we see this as a problem that could benefit from an algorithmic decision-making process that aims to be equitable. In this paper, we first explore zoning classification as a supervised learning task based on existing zoning data in a locality. We extensively collect publicly available data from different sources to train models on them and show how rather complex models are needed to learn to predict the classification accurately. Furthermore, we do a counterfactual analysis based on socioeconomic features to show how they seem to be important indicators for the classification problem in currently available data. We hope that our exploratory paper will lead to more work in applying computational techniques for fair zoning assignments.

Submission Number: 14