Identifying representative weeks: A clustering analysis of urban dynamics based on public transport data

Abstract

Transportation is key to understanding urban dynamics, with public transport bus systems shaping mobility patterns, accessibility, and activity, especially in developing countries where this mode accounts for a significant share of trips. Passive data like GPS and smart card records can reveal these patterns when properly processed. This study proposes a clustering-based methodology to analyze public transport bus data collected over an arbitrarily long period (at least a week) to identify time blocks with similar dynamics without predefined structures. If the dynamics for each time block in a week is the same over time, a representative week can be constructed, showing the most likely dynamics for each time block in a week. The analysis considers three dimensions: demand (passenger counts), supply (distance traveled by buses), and level of service (bus speeds). Cluster results generate a representative week in terms of mobility indicators and transport operations, enabling analysis and comparison of dynamics across different city zones. Using data from Santiago de Chile's bus system for August 2019 and April 2020, the methodology was applied to 10 city zones. Results highlighted distinct dynamics across zones and the need to incorporate all three dimensions for representative weeks. Regular application of this approach is crucial, as cluster characteristics evolve over time. While promising venues for future development remain, our methodology provides a flexible, robust data-driven foundation for understanding urban transport dynamics, adaptable to different cities and supporting evidence-based decisions.