From Data to Design: Using mobile app data and machine learning techniques for street-level walkability assessment

Abstract

Recognizing streets as social spaces, urban planners in recent years aim to reposition pedestrians at the forefront of city design, challenging decades of car-centric urban planning (Salazar Miranda et al., 2021). One result is the increasing interest in the interplay between walking and the built environment among scholars and practitioners striving to foster sustainable transportation and cultivate healthy urban communities globally (Yencha, 2019). However, while pedestrians interact closely with the street environment, walkability assessments to this date have predominantly operated at the neighbourhood level, neglecting essential human-scale characteristics. Moreover, the exploration of walkability and the built environment has traditionally relied on constrained and temporally confined methods, such as surveys and observations (Ferrer & Ruiz, 2018). To bridge these research gaps and gain fresh insights into the intricate connection between walking and the built environment, this study employs machine learning techniques to scrutinize mobile app data capturing pedestrian traffic, coupled with street characteristics examination.

Specifically, the study leverages tree-based algorithms to unravel the nuanced relationship between pedestrian volume and various built environment features at the street level, spanning diverse time periods. Pedestrian traffic data from Tel Aviv, Israel, forms the basis of the analysis, with meticulous consideration given to seasonal variations, days of the week, and hours of the day. The investigation spans an extensive dataset of over 8,000 street segments and encompasses 20 street-level characteristics, including elements such as trees, lighting, bus stations, businesses, educational and cultural institutions, health services, and residential density, among others, providing a robust foundation for evaluating the dynamic association between pedestrian volume and urban features.

The study's outcomes yield fresh perspectives on the relative significance of diverse characteristics in influencing walking activity, elucidating distinct street profiles associated with varying levels of pedestrian volumes. Notably, our findings underscore the crucial role of time variables in the discussion of the relationship between walking and street characteristics. The model results reveal that the importance attributed to street features varies across different time definitions, shedding light on the temporal dynamics of pedestrian movement. Further, the importance of street-level walkability assessment is emphasized in the results. for example, land-use mix, which is commonly attributed to more walking at neighbourhood-level, was found with low significance at street level; Areas with low walkability overall also include street segments with very high walking volume and vice versa.

This research carries implications for decision-makers and urban planners. By unveiling pedestrians' behaviours and preferences at the street level, the study suggests that the use of bigdata technologies and artificial intelligence analysis can equip stakeholders with valuable insights to guide more efficient infrastructure investments and informed planning decisions. The nuanced understanding of the relationship between walking and the built environment, facilitated by machine learning methodologies, offers a paradigm shift in approaching urban planning and walkability research, encouraging the creation of environments that not only promote active transportation but also contribute to the establishment of healthier and more liveable communities.