From Intuition to Reasoning: Analyzing Correlative Attributes of Walkability in Urban Environments with Machine Learning

Abstract

A fundamental challenge in the planning and operation of modern cities is their walkability. Walkability is typically assessed using geo-information system (GIS) or real-time observations. These existing methods, however, are not suited to the task in several key aspects. GIS-based assessment is inherently limited in capturing the details of a space, and observation-based methods are time and resource consuming. To overcome these limitations, we introduce a novel machine learning (ML) based approach. Our main concept is to make walkability an ML problem, where sites or locations are defined as data points. The data points are characterised by features extracted from street images. The ultimate quantity of interesting aspects (or labels) of a data point determines its level of walkability. Our assessment of walkability is based on the perceived accessibility of sites as measured via survey. Roughly speaking, our ML approach learns correlations between the presence of specific objects such as trees, buildings, sidewalks, and the perceived walkability of a specific location. The main methodological contribution of our research is a novel feature extraction method based on semantic segmentation techniques. The extracted features are fed into different off-the-shelf supervised ML methods and compared. The results demonstrate the usefulness of our approach to predict the walkability of an urban location based on an ML analysis of street image content.