Clustering Analysis of the Spatio-Temporal On-Street Parking Occupancy Data: A Case Study in Hong Kong

Fan Wu, Wei MaSustainability 2022 · DOI

TL;DR

This paper analyzes real-time on-street parking occupancy data from Hong Kong to uncover distinct spatio-temporal parking patterns. By integrating t-SNE for visualization and k-means for clustering, the study reveals that Hong Kong's on-street parking is highly utilized (over 80% occupancy) and exhibits three consistent patterns, which can be linked to land-use factors, offering valuable insights for smart city management.

What Problem Does This Paper Address?

Urban metropolises worldwide grapple with the challenge of on-street parking, which significantly contributes to traffic congestion due to drivers circling for available spots. Hong Kong, a densely populated global hub, faces particularly high parking demand. While the Hong Kong government has deployed new smart parking meters to collect real-time occupancy data, there's a lack of research analyzing this data to understand the unique parking behaviors and patterns within the city. This gap hinders the development of effective, data-driven smart parking management strategies.

Methods

The study employs a two-pronged approach combining dimension reduction and clustering:

  • t-Distributed Stochastic Neighbor Embedding (t-SNE): This technique is used for visualizing high-dimensional spatio-temporal parking data in a low-dimensional (2D) space. t-SNE is chosen for its effectiveness in preserving local data structure and separating dissimilar data points, making it suitable for revealing underlying patterns in complex datasets.
  • k-Means Clustering: After dimension reduction, k-means clustering is applied to group similar parking patterns. This unsupervised learning algorithm partitions the data into a predefined number of clusters (k), allowing for the identification of distinct temporal and spatial parking behaviors across different districts and times. The elbow method was used to determine the optimal number of clusters.

Datasets

The primary dataset used in this study is real-time on-street parking occupancy data collected from new parking meters in Hong Kong during July 2021. This data was obtained from DATA.GOV.HK and includes information such as parking meter ID, occupancy status, and timestamp, collected every five minutes. The study also references population distribution data and a distribution map of entertainment venues in Hong Kong to help explain observed parking patterns.

Key Insights and Ideas

  • Ubiquitous High Occupancy: The most striking insight is the consistently high average on-street parking occupancy rate in Hong Kong, often exceeding 80% throughout the day. This highlights the intense demand for on-street parking in the city.
  • Distinct Spatio-Temporal Patterns: Despite the overall high occupancy, the analysis reveals three consistent spatio-temporal parking patterns across different districts. This suggests that parking demand isn't uniform and varies significantly based on location and time.
  • Weekend vs. Weekday Dynamics: The study confirms that parking demand is generally higher on weekends than on weekdays, with noticeable differences in peak times and duration. This aligns with typical urban behavior where leisure activities increase on weekends.
  • Land Use as a Predictor: A key finding is that these identified parking patterns can be explained by land-use factors, such as the density of entertainment venues. This provides a crucial link between urban planning elements and parking behavior.
  • Unique Nighttime Demand: Interestingly, one of the identified parking patterns shows higher demand at night, which the authors suggest might be a unique characteristic of Hong Kong due to limited land supply and potentially different urban activity patterns.

Key Results

  • The average on-street parking occupancy rate in Hong Kong is consistently above 80% throughout the day.
  • Three distinct on-street parking patterns were identified through t-SNE visualization and k-means clustering.
  • Parking demand is generally higher on weekends compared to weekdays.
  • The study identified specific districts with higher parking demand during the day versus at night, and others with more stable occupancy throughout the day.
  • The clustering analysis identified two temporal patterns (weekday vs. weekend/holiday) and four spatial patterns for the 18 districts.

Limitations and Future Work

The authors acknowledge several limitations:

  • Data Period: The study utilized data from only one month (July 2021), which might not fully capture long-term seasonal variations or trends, especially since the new parking meters were still being fully installed. Future studies should incorporate longer time horizons and data from more meters.
  • Granularity: While the analysis covers districts, a more granular study at the block or road level could provide deeper insights.
  • Sensor Failures: The impact of sensor failures on data accuracy was not thoroughly studied and warrants further investigation.
  • Theoretical Modeling: The current study is data-driven. Future work could involve developing theoretical models to better depict the complex interactions within parking systems.

Despite these limitations, the proposed methodology of integrating t-SNE and k-means for spatio-temporal parking analysis is presented as a generalized approach applicable to other cities with similar smart parking data.

Summary updated

Abstract

Parking plays an essential role in urban mobility systems across the globe, especially in metropolises. Hong Kong is a global financial center, international shipping hub, fast-growing tourism city, and major aviation hub, and it thus has a high demand for parking. As one of the initiatives for smart city development, the Hong Kong government has already taken action to install new on-street parking meters and release real-time parking occupancy information to the public. The data have been released for months, yet, to the best of our knowledge, there has been no study analyzing the data and identifying their unique characteristics for Hong Kong. In view of this, we examined the spatio-temporal patterns of on-street parking in Hong Kong using the data from the new meters. We integrate the t-SNE and k-means methods to simultaneously visualize and cluster the parking occupancy data. We found that the average on-street parking occupancy in Hong Kong is over 80% throughout the day, and three parking patterns are consistently identified by direct data visualization and clustering results. Additionally, the parking patterns in Hong Kong can be explained using land-use factors. Overall, this study can help the government better understand the unique characteristics of on-street parking and develop smart management strategies for Hong Kong.
BibTeX
@inproceedings{wu2022clustering,
  author = {Fan Wu and Wei Ma},
  title = {{Clustering Analysis of the Spatio-Temporal On-Street Parking Occupancy Data: A Case Study in Hong Kong}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su14137957},
}
Figure