Mining individual daily commuting patterns of dockless bike-sharing users: a two-layer framework integrating spatiotemporal flow clustering and rule-based decision trees

Overview

• Examines the use of dockless bike-sharing data for studying urban transportation and travel behavior
• Proposes a two-layer framework to mine individual cyclists' daily home-work commuting patterns from bike-sharing trip data
• Demonstrates the effectiveness of the framework using over 200 million dockless bike-sharing trip records in Shenzhen

Plain English Explanation

Dockless bike-sharing systems have become increasingly popular, leading to a growing interest in using this data for urban transportation research. However, few studies have focused on understanding the daily mobility patterns of individual cyclists, which is important for improving urban transportation planning.

This study presents a two-part framework to address this gap. The first part uses improved flow clustering methods to group similar bike-sharing trips together. The second part applies multiple decision trees to analyze these trip patterns and identify common commuting behaviors, such as biking only versus biking combined with public transit.

The researchers tested this framework using over 200 million bike-sharing trip records from Shenzhen, China. They found that about 74% of commuters only use bike-sharing, while 26% combine it with other modes of transportation, like buses or trains. Interestingly, many bike-sharing commuters live in areas with lower housing costs, often near older neighborhoods or industrial parks. The study also revealed that biking-only commuters tend to work longer hours compared to those who combine biking with transit.

Overall, this framework provides a cost-effective way to better understand the daily mobility patterns of urban residents, which can inform more effective cycling infrastructure and transportation planning.

Technical Explanation

The researchers developed a two-layer framework to mine individual cyclists' daily home-work commuting patterns from dockless bike-sharing data. The first layer uses an improved flow clustering method to group similar bike-sharing trips based on their origin, destination, and travel time. This helps identify common commuting routes and patterns.

The second layer then applies multiple rule-based decision trees to analyze the clustered trip data and classify commuters into two main categories: "Only-biking commuters" and "Biking-with-transit commuters". This allows the researchers to gain detailed insights into the daily mobility behaviors of different types of bike-sharing users.

The framework was tested using over 200 million dockless bike-sharing trip records from Shenzhen, China. The results show that 74.38% of commuters only use bike-sharing, while 25.62% combine it with other modes of transportation. Further analysis revealed that many bike-sharing commuters live in areas with lower housing costs, such as urban villages and old communities near the city center. Additionally, the "Only-biking commuters" tend to have a higher proportion of overtime work compared to "Biking-with-transit commuters", with the Longhua Industrial Park area having the longest average working hours of over 10 hours per day.

Critical Analysis

The study presents a comprehensive framework for mining individual cyclists' daily commuting patterns from dockless bike-sharing data, which is a significant contribution to the field of urban transportation research. The use of improved flow clustering and rule-based decision trees allows for a more nuanced understanding of commuting behaviors compared to previous studies that focused on aggregate-level analysis.

However, the study acknowledges several limitations. First, the framework relies on the availability of user IDs in the bike-sharing data, which may not always be the case. Second, the decision tree models used to classify commuters into different categories may not capture the full complexity of individual travel behaviors. There could be additional factors, such as demographic and socioeconomic characteristics, that influence commuting patterns.

Future research could explore ways to overcome these limitations, such as leveraging cellular network data or predictive maintenance models to supplement the bike-sharing data. Additionally, the framework could be tested in other cities to assess its generalizability and identify any context-specific factors that may influence commuting patterns.

Conclusion

This study presents a comprehensive framework for mining individual cyclists' daily home-work commuting patterns from dockless bike-sharing data. The findings provide valuable insights into the travel behaviors of different types of bike-sharing users, such as the prevalence of "Only-biking commuters" and their tendency to work longer hours.

The proposed framework offers a cost-effective way to understand residents' non-motorized mobility patterns, which can inform more effective urban transportation planning and policies. By considering the nuances of individual travel behaviors, this research paves the way for subsequent studies on fine-scale cycling patterns and their implications for sustainable urban development.