Discovering Urban Spatial-Temporal
Structure from Human Activity Patterns
Shan Jiang, [email protected]
Joseph Ferreira, Jr., [email protected]
Marta C. Gonzalez, [email protected]
ACM SIGKDD International Workshop on Urban Computing (UrbComp 2012)
August 12, 2012 - Beijing, China
Outline
1. Introduction
Motivation | Data & Study Area
2. Urban spatial-temporal structure(STS)
Measurement | The Chicago Case
3. Clustering daily activity patterns and traces Activity Profile Clusters | Trace Clusters
4. Urban spatial-temporal structure by region, activity pattern and type
5. Conclusions
1. Introduction Background • Cities have evolved from monocentric to polycentric forms,
due to – the improvement of transportation systems – the rise of consumer city (Glaeser et al. 2001)
• Swelling cities have become data repositories of human activities, due to – the emergence of the use of Information and Communication
Technologies (ICT) in everyday life – ubiquitous urban sensors (e.g., GPS, mobile phone, online social media)
Challenges • Traditional measurement of urban structure
– measured by population and employment density, is static – cannot capture the dynamics in space and time in cities
• Urban sensing data – little information on social demographics/activity types of the users is known
to researchers due to legal/privacy constraints
1. Introduction Our Approach
• We expand the traditional understanding of urban structure from spatial dimension to spatial-temporal dimensions
• We detect clusters of individuals by daily activity patterns, integrated with their usage of space and time, and show that daily routines can be highly predictable, with clear differences depending on the group.
Data & Study Area
• Travel survey collected by the Metropolitan Planning Organization
– are representative of the population – can inform us about “who, what, when, where, why and how of travel for each
person in a surveyed household”
• In this study, we use the 2008 Chicago Travel Tracker survey data.
– a 1-day or 2-day survey, including a total of 10,552 households, 30,000+ individuals. – We use Monday to Thursday as a representative weekday sample (23,527 distinct
individuals)
Study Area & Data Chicago Temporal Activity Patterns: Weekday
5 S. Jiang, J. Ferreira, M. Gonzalez (2012)
6 Ref: Jiang, S., J. Ferreira & M. González (2012) Clustering daily patterns of human activities in the city. Data Mining and Knowledge Discovery, 25, 478-510.
Chicago Temporal Activity Patterns: Weekday
2. Urban Spatial-Temporal Structure 2.1 Measurement and Estimation
• We define a spatial-temporal space S as:
• Spatial-Temporal Activity Density
• Time-cumulative Spatial Activity Density
• Kernel Density Estimator
[1]
[2]
[3]
[4]
2. Urban Spatial-Temporal Structure 2.2 Chicago Metro-Area Example
Home Work School
Recreation/ entertainment
Shopping / errands
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
104
207
311
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
980
1959
2939
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
245
490
735
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
428
856
1284
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
2602
5204
7806
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
1020
2041
3061
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
1407
2815
4222
Time of Day (in Hour)
Sam
p. #
4 6 8 10 12 14 16 18 20 22 24 2
1056
2113
3169
Home
Work
Schl.
Trans.
Shopping
Personal
Rec.
Civic
Other
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
% o
f V
ol.
4 8 12 16 20 24 0.0
33.3
66.7
100.0
Time of Day (in Hour)
% o
f V
ol.
4 8 12 16 20 24 0.0
9.0
18.1
27.1
% o
f V
ol.
4 8 12 16 20 24 0.0
4.4
8.7
13.1
% o
f V
ol.
4 8 12 16 20 24 0.0
2.3
4.5
6.8
% o
f V
ol.
4 8 12 16 20 24 0.0
3.2
6.4
9.6
% o
f V
ol.
4 8 12 16 20 24 0.0
1.8
3.6
5.5
% o
f V
ol.
4 8 12 16 20 24 0.0
6.6
13.1
19.7
% o
f V
ol.
4 8 12 16 20 24 0.0
4.2
8.4
12.6
% o
f V
ol.
4 8 12 16 20 24 0.0
2.9
5.8
8.6
Time of Day (in Hour)
% o
f V
ol.
Cluster # 2 (1.3%)
Cluster # 8 (12.5%)
Cluster # 7 (3.1%)
Cluster # 4 (5.5%)
Cluster # 6 (33.2%)
Cluster # 3 (13.0%)
Cluster # 5 (17.9%)
Cluster # 1 (13.5%)
3. Clustering Activities & Traces 3.1 Activity Profile Clusters
Overnight adventurers
Students
Afternoon workers
Afternoon adventurers
Stay-at-home
Morning adventurers
Regular workers
Early-bird workers
Ref.: Jiang, S., J. Ferreira & M. González (2012) Clustering daily patterns of human activities in the city. Data Mining and Knowledge Discovery, 25, 478-510.
3. Clustering Activities & Traces 3.2 Trace Clusters
Cluster numbers (k) and the Dunn’s Index (DI)
4. Urban spatial-temporal structure by region, activity profile and activity type
5. Conclusions To facilitate urban planners and scholars to understand
– the dynamics and complexity of polycentric cities, and – how cities have been utilized by different types of individuals for
different activity types in space and time …
We propose a concept of urban spatial-temporal structure(STS) – Measurement: time-cumulative spatial activity density – Estimation: kernel density estimator
We analyzed the STS of a polycentric metropolitan area – by clustering individuals by activity patterns and traces using k-means
algorithm via PCA , and – by estimating and visualizing the time-cumulative spatial densities of
various activities by person types (of activity profiles and trace clusters) in one of the regions of Chicago
This analysis presents the basis to capture collective activities at large scales and expand our perception of urban structure from the spatial dimension to spatial-temporal dimension.
Acknowledgements
This research was funded in part by the MIT Department of Urban Studies and Planning, by the US Department of Transportation Region One University Transportation Center, and by the Singapore National Research Foundation (NRF) through the Singapore-MIT Alliance for Research and Technology (SMART) Center for Future Mobility (FM).
THANK YOU! Questions? Email: [email protected]