ChapterPDF Available

A Clustering-Based Framework for Understanding Individuals’ Travel Mode Choice Behavior

Authors:

Abstract

Travel mode choice analysis is a central aspect of understanding human mobility and plays an important role in urban transportation and planning. The emergence of passively recorded movement data with spatio-temporal and semantic information offers opportunities for uncovering individuals’ travel mode choice behavior. Considering that many of these choices are highly regular and are performed in similar manners by different groups of people, it is desirable to identify these groups and their characteristic behavior (e.g. for educational or political incentives or to find environmentally-friendly people). Previous research mainly grouped people according to “mobility snapshots”, i.e. mobility patterns exhibited at a single point in time. We argue that especially when considering the change of behavior over time, we need to investigate the behavioral dynamic processes resp. the change of travel mode choices over time. We present a framework that can be used to cluster people according to the dynamics of their travel mode choice behavior, based on automatically tracked GPS data. We test the framework on a large user sample of 107 persons in Switzerland and interpret their travel mode choice behavior patterns based on the clustering results. This facilitates understanding people’s travel mode choice behavior in multimodal transportation and how to design reasonable alternatives to private cars for more sustainable cities.
Metadata of the chapter that will be visualized in
SpringerLink
Book Title Geospatial Technologies for Local and Regional Development
Series Title
Chapter Title A Clustering-Based Framework for Understanding Individuals’ Travel Mode Choice Behavior
Copyright Year 2020
Copyright HolderName Springer Nature Switzerland AG
Corresponding Author Family Name Zhao
Particle
Given Name Pengxiang
Prefix
Suffix
Role
Division
Organization Institute of Cartography and Geoinformation, ETH Zurich
Address Stefano-Franscini-Platz 5, 8093, Zurich, Switzerland
Email pezhao@ethz.ch
Author Family Name Bucher
Particle
Given Name Dominik
Prefix
Suffix
Role
Division
Organization Institute of Cartography and Geoinformation, ETH Zurich
Address Stefano-Franscini-Platz 5, 8093, Zurich, Switzerland
Email
Author Family Name Martin
Particle
Given Name Henry
Prefix
Suffix
Role
Division
Organization Institute of Cartography and Geoinformation, ETH Zurich
Address Stefano-Franscini-Platz 5, 8093, Zurich, Switzerland
Email
Author Family Name Raubal
Particle
Given Name Martin
Prefix
Suffix
Role
Division
Organization Institute of Cartography and Geoinformation, ETH Zurich
Address Stefano-Franscini-Platz 5, 8093, Zurich, Switzerland
Email
Abstract Travel mode choice analysis is a central aspect of understanding human mobility and plays an important
role in urban transportation and planning. The emergence of passively recorded movement data with
spatio-temporal and semantic information offers opportunities for uncovering individuals’ travel mode
choice behavior. Considering that many of these choices are highly regular and are performed in similar
manners by different groups of people, it is desirable to identify these groups and their characteristic
behavior (e.g. for educational or political incentives or to find environmentally-friendly people). Previous
research mainly grouped people according to “mobility snapshots”, i.e. mobility patterns exhibited at a
single point in time. We argue that especially when considering the change of behavior over time, we need
to investigate the behavioral dynamic processes resp. the change of travel mode choices over time. We
present a framework that can be used to cluster people according to the dynamics of their travel mode
choice behavior, based on automatically tracked GPS data. We test the framework on a large user sample
of 107 persons in Switzerland and interpret their travel mode choice behavior patterns based on the
clustering results. This facilitates understanding people’s travel mode choice behavior in multimodal
transportation and how to design reasonable alternatives to private cars for more sustainable cities.
Keywords
(separated by '-')
Human movement data - Travel mode choice behavior - Autocorrelation - Hierarchical clustering
UNCORRECTED PROOF
A Clustering-Based Framework for
Understanding Individuals’ Travel Mode
Choice Behavior
Pengxiang Zhao, Dominik Bucher, Henry Martin and Martin Raubal
Abstract Travel mode choice analysis is a central aspect of understanding human
1
mobility and plays an important role in urban transportation and planning. The emer-2
gence of passively recorded movement data with spatio-temporal and semantic infor-3
mation offers opportunities for uncovering individuals’ travel mode choice behavior.4
Considering that many of these choices are highly regular and are performed in sim-5
ilar manners by different groups of people, it is desirable to identify these groups6
and their characteristic behavior (e.g. for educational or political incentives or to find7
environmentally-friendly people). Previous research mainly grouped people accord-8
ing to “mobility snapshots”, i.e. mobility patterns exhibited at a single point in time.9
We argue that especially when considering the change of behavior over time, we10
need to investigate the behavioral dynamic processes resp. the change of travel mode11
choices over time. We present a framework that can be used to cluster people accord-12
ing to the dynamics of their travel mode choice behavior, based on automatically13
tracked GPS data. We test the framework on a large user sample of 107 persons in14
Switzerland and interpret their travel mode choice behavior patterns based on the15
clustering results. This facilitates understanding people’s travel mode choice behav-16
ior in multimodal transportation and how to design reasonable alternatives to private17
cars for more sustainable cities.18
Keywords Human movement data ·Travel mode choice behavior ·19
Autocorrelation ·Hierarchical clustering20
1 Introduction21
As one of the environmentally relevant behaviors, travel mode choice has become22
increasingly important with the rising social concern for the environment (Hunecke23
P. Zhao (B)·D. Bucher ·H. Martin ·M. Raubal
Institute of Cartography and Geoinformation, ETH Zurich,
Stefano-Franscini-Platz 5, 8093 Zurich, Switzerland
e-mail: pezhao@ethz.ch
© Springer Nature Switzerland AG 2020
P. Kyriakidis et al. (eds.), Geospatial Technologies for Local and Regional
Development, Lecture Notes in Geoinformation and Cartography,
https://doi.org/10.1007/978-3-030-14745- 7_5
1
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
2P. Zhao et al.
et al. 2001). Specifically, traffic-related carbon dioxide (CO2) emissions resulting24
from cars and taxis speed up the greenhouse effect and pose a large threat to the25
environment (Zhao et al. 2017a). When trying to provide sustainable travel modes,26
policy-makers and mobility providers are facing both challenges and opportunities.27
In recent years, multimodality has already been considered for a more sustainable28
future urban mobility, as it offers an attractive alternative to the private car through29
combinations of (more ecological) travel modes such as public transport, electric30
car (e-car) and bicycle (Klinger 2017). Especially, the proliferation of Mobility as a31
Service (MaaS) makes it more convenient for people to book a delivery service with a32
range of travel modes. Compared to private cars and taxis, the use of public transport33
and e-cars is a more environmentally friendly type of travel behavior due to their34
lower CO2emissions. Therefore, travel mode choice has always been a continuing35
research topic in the fields of Geographic Information Science and transportation36
planning.37
Over the past decades, the vast majority of studies on travel mode choice have38
concentrated on modeling and analyzing the related influencing factors from travel39
surveys and questionnaires (Chen et al. 2008; Murtagh et al. 2012;Anetal.2015;40
Daisy et al. 2018). However, less emphasis has been given to explore individu-41
als’ travel mode choice behavior patterns. Although examining the influence factors42
of travel mode choice is necessary, understanding individuals’ travel mode choice43
behavior patterns is also significant for urban transportation planning. For instance,44
it is meaningful to discover whether there is a large group of people who choose45
the car as sole travel mode or if there are people who usually choose a combination46
of travel modes (Heinen and Chatterjee 2015). Due to the high cost and large time47
requirements, travel survey data normally merely record the respondents’ status quo48
of travel mode choice behavior, thereby ignoring the behavioral change processes.49
Hence, these datasets fail to uncover and analyze interpersonal and intrapersonal50
travel mode choice behavior patterns.51
The emergence and prevalence of GPS-based human movement data (e.g. mobile52
phone records, taxi trajectory data) facilitates characterizing and analyzing human53
mobility patterns at the aggregate and individual levels (Bucher et al. 2019; Yuan and54
Raubal 2014; Zhao et al. 2017b; Jonietz and Bucher 2018). Undoubtedly, there are55
numerous studies which investigate human travel behavior patterns from movement56
trajectory data. For instance, Barbosa et al. (2018) reviewed research on reproducing57
human mobility patterns from various movement data sources in recent years, which58
shed light on the fundamental modeling approaches and technical methods of human59
mobility. However, there has not been sufficient research on exploring and under-60
standing human travel mode choice patterns at the individual level. The objective of61
this work is to understand individuals’ travel mode choice behavior patterns through62
categorizing users with similar mode choice behavior patterns based on their trajec-63
tory data. Exploring these patterns will be helpful for policy-makers to understand64
people’s travel mode choice behavior, and design and implement more sustainable65
mobility strategies.66
In this study, we propose a clustering-based framework for understanding indi-67
viduals’ travel mode choice behavior patterns. Specifically, the framework contains68
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 3
three main steps: (1) construct features in terms of time series to describe individuals’69
travel mode choice behavior; (2) measure the similarity of different users based on70
autocorrelation coefficients; (3) segregate individuals into groups based on their sim-71
ilarity using hierarchical clustering. The reason for choosing hierarchical clustering72
is that researchers can define their own similarity or distance measures to generate73
a similarity matrix according to their research purpose and because it can be con-74
veniently visualized using a dendrogram. The experimental dataset comes from the75
large-scale pilot study SBB Green Class conducted in Switzerland, which will be76
introduced in Sect. 3.77
The article is organized as follows: Related work regarding travel mode choice78
and time series clustering are reviewed in Sect.2. Section 3describes the dataset79
used in this study. Section4presents the overall framework for understanding travel80
mode choice behavior patterns. The experimental results are shown in Sect.5and we81
discuss and conclude this research in Sect. 6.82
2 Related Work83
2.1 Examining Individuals’ Travel Mode Choice84
Travel mode choice has been investigated based on choice behavior theory for several85
decades (Glasser 1999). As mentioned in the introduction, the central issue of the86
related studies is to model and analyze the relationship between people’s travel mode87
choices and their influencing factors. It has been demonstrated that travel mode choice88
is impacted by a variety of factors, such as built environment (Chen et al. 2008;Ding89
et al. 2017; Sun et al. 2017; Han et al. 2018), individual characteristics (Murtagh et al.90
2012; Vij et al. 2013; Böcker et al. 2017), weather conditions (Böcker et al. 2013;Liu91
et al. 2015), or travel time and distance (An et al. 2015; Daisy et al. 2018). However,92
little attention has been paid to studies on detecting individuals’ travel mode choice93
behavior patterns. Although there are several studies that investigate travelers’ mode94
choice behavior by grouping individuals, they normally divide the travelers into95
groups according to socioeconomic information and neglect the behavioral change96
processes of travel mode choice (Ding and Zhang 2016).97
With the widespread usage of smart phones and location-based services (Huang98
et al. 2018), it has become convenient to record individuals’ daily travel activities99
over longer periods. The emergence of GPS-based human movement data has spurred100
numerous studies on individuals’ travel behavior patterns from their trajectories.101
For instance, Shen and Cheng (2016) proposed a theoretical framework to divide102
users into subgroups according to their travel behavior patterns using individual103
trajectory data. An integrated framework was developed to analyze human mobility104
patterns from volunteered GPS trajectories and contextual information. Specifically,105
how individuals’ travel mode choice depends on their residential location, age or106
gender (Siła-Nowicka et al. 2016). The individual travel behavior regularity was107
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
4P. Zhao et al.
investigated through constructing a travel behavior graph model. The graph can be108
used to indicate an individual’s travel routes and forecast travel mode choice behavior109
(Liang et al. 2018). Several studies specifically considered how human movement110
data in combination with location-aware information technology can be used to111
influence mobility behavior (Froehlich et al. 2009; Bucher et al. 2016). This line112
of research is often driven by the desire to lower greenhouse gas emissions and113
make mobility more sustainable (Weiser et al. 2016). In this study, we focus on114
understanding individuals’ travel mode choice behavior patterns, more specifically115
on the temporal patterns of individuals’ travel duration and distance by means of116
each type of travel mode, which reflects the behavioral change processes of travel117
mode choice. Autocorrelation is employed to measure the similarity of time series,118
as it allows capturing regularities inherent in some behavior, and can be used to119
effectively calculate the similarity of short time series (Aghabozorgi et al. 2015).120
The background of clustering time series based on their similarity is presented in121
Sect. 2.2.122
2.2 Clustering Time Series Based on their Similarity123
Clustering time series datasets has been universally done in diverse scientific disci-124
plines and domains. It facilitates researchers and data analysts to discover valuable125
information and knowledge in an unsupervised way. Measuring the similarity of126
time series has always been an important research question regarding time series127
clustering (Gunopulos and Das 2001). The traditional method to measure the simi-128
larity of time series is the Euclidean distance (ED), which regards time series as a129
vector across all time points and then calculates the sum of the distances between130
corresponding points. In recent years, a growing body of measurement methods has131
emerged to calculate the similarity of time series.132
For time series clustering, similarity measurement approaches of time series data133
were summarized into four categories according to the characteristics of the time134
series, namely shape-based,compression-based,model-based and feature-based135
approaches (Aghabozorgi et al. 2015). Shape-based similarity measures are essen-136
tially used to discover similar time series in shape and time, and include Dynamic137
Time Warping (DTW), or Longest Common Subsequence (LCSS) (Górecki 2018).138
As a classical shape-based similarity measurement, DTW has been widely utilized in139
time series clustering. For instance, Yuan and Raubal (2012) explored dynamic urban140
mobility patterns from mobile phone data by measuring the similarity of time series141
representing the dynamic mobility patterns of different urban areas. Subsequently, a142
large number of improved DTW methods was proposed to measure the similarity of143
time series (Bankó and Abonyi 2012; Łuczak 2016;Yeetal.2017). Compression-144
based similarity is applicable in short and long time series, and includes Autocorre-145
lation or Pearson’s correlation coefficient. For example, Yue et al. (2018) developed146
a spectral clustering framework to understand the intertwined usage of bus, metro,147
and taxi in urban space. Specifically, the similarity of time series that represents the148
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 5
ridership patterns of mass transit modes in different urban spaces was measured by149
calculating autocorrelation coefficients. Model-based and feature-based measures150
are suitable for long time series, and include ARMA models (Xiong and Yeung151
2002), Hidden Markov Models (HMM) (Dias et al. 2015), Fourier transformation152
(Gao et al. 2015), and wavelet transformation (Barragan et al. 2016).153
Since the time series in this work are generated from daily aggregates and154
are thus rather short, we confine the similarity measurements to shape-based and155
compression-based approaches. Compared with autocorrelation, the DTW algorithm156
neglects the interior correlation within the time series. Considering the regularity and157
periodicity of individuals’ travel behavior patterns, an autocorrelation approach is158
chosen to measure the similarity between the time series, which will be introduced159
in Sect. 4.3 in detail. Although autocorrelation has been employed to measure the160
similarity of time series, to the best of our knowledge no studies have applied it to161
explore individuals’ travel mode choice behavior patterns.162
3 Data163
This study is based on the dataset from the large-scale pilot project SBB Green Class164
in Switzerland. SBB Green Class was carried out by the Swiss Federal Railways165
(SBB) and offered 139 participants a Mobility as a Service (MaaS) package, which166
included a general public transport pass, a BMW i3 electric vehicle (with a corre-167
sponding charging station at home), memberships to common car- and bikesharing168
programs, as well as a parking spot at a train station of the participant’s choice.169
While the participants were primarily selected based on their geographic location,170
the (financial) participation preconditions lead to a bias towards middle and upper171
class people. As part of the pilot, the participants were asked to install a commercial172
application on their smart phones to track their daily movement. The recorded GPS173
data (approx. one GPS recording every 1–5 min, with a spatial accuracy in the order174
of tens of meters) were automatically segmented into trajectories and stay points by175
the app. Next to the raw spatial and temporal information, the data include semantic176
information about the travel mode of each trip and the purposes of stay points, which177
have been manually validated by the users themselves (the transport modes proposed178
for validation were identified by the commercial tracking app, but likely to be based179
on accelerometer data). The validated travel modes contain airplane,bicycle,boat,180
bus,car,coach,e-bicycle,e-car,train,tram and walk. While the whole project ran181
from January to December 2017, we here select a subset of 183 days, covering April182
to September 2017. In addition, selecting April 2017 as a starting date for this study183
allowed people at least one month’s time to explore the new mobility options and184
settle for a regular behavior (i.e. the novelty of the MaaS offer and especially the185
electric car has worn off).186
Since this study focuses on individuals’ daily travel trips, we exclude airplane187
trips. Additionally, in such a practical application, GPS trajectories are not perfect188
due to various influencing factors, and people have gaps in their recordings (e.g.,189
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
6P. Zhao et al.
due to a lack of battery or voluntarily turning off the tracking device). Therefore, we190
filter the dataset based on the available data to ensure the trajectories are reliable. As191
a filtering criteria, we remove all users that either have a gap of more than 30 days192
duration or who recorded data on less than 150 days. This leaves us with 107 active193
users with mostly continuous records which are further used within this study.194
4Method195
In this study, a clustering-based framework is proposed to understand individuals’196
travel mode choice behavior using GPS trajectory data. The overall framework con-197
sists of five steps, which are shown in Fig.1. First, we construct features to represent198
individual travel mode choice behavior. The second step is to interpolate the gaps199
for the days of data loss. Third, we measure the similarity of individual travel mode200
choice behaviors based on autocorrelation. The fourth step is to divide the individuals201
into different groups by means of a hierarchical clustering algorithm. The last step202
is to conduct behavior pattern analysis based on the clustering results.203
4.1 Constructing Features of Travel Mode Choice204
In this section, we aim to extract a series of descriptive features of the individuals’205
travel mode choice behavior, namely the modal split during the same period (Jonietz206
et al. 2018). For instance, how long (i.e. duration) or how far (i.e. distance) someone207
utilizes various travel modes every day. On this basis, we construct duration-based208
and distance-based features respectively to depict the travel mode choice behavior209
for each user, which capture the temporal fluctuations of duration and distance by210
means of a certain travel mode during a period. In this study, one day is selected as211
the temporal granularity. Additionally, we only consider days were we have tracked212
over 70% of a users day and treat all other days as missing values.213
For an individual’s travel duration and distance by means of a certain type of214
travel mode, the features Tiand Dican be denoted as 1 ×183 (days) vectors:215
Ti=[t1
i,t2
i,...,t183
i](1)216
Fig. 1 Workflow of the framework: the raw trajectory data are processed in five consecutive steps
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 7
Di=[d1
i,d2
i,...,d183
i](2)217
where i(1,2,...,10), which corresponds to ten types of travel modes (i.e. bicycle,218
boat, bus, car, coach, e-bicycle, e-car, train, tram and walk). tj
iand dj
irepresent the219
duration and distance of the ith type of travel mode on the jth day, respectively.220
Ultimately, duration-based and distance-based features are obtained, which are221
expressed as 10 time series respectively. Figure 2visualizes the constructed features222
of two exemplary users based on the above-mentioned 10 types of travel modes.223
As can be seen in the features, the two users exhibit different travel mode choice224
behavior patterns. The first user (red) selects multiple travel modes for daily trips,225
including car, e-car, tram and walk, while the second user (blue) mainly depends on226
e-car and walk as well as car for long-distance trips.227
4.2 Interpolating the Gaps228
Since significant proportions of human activities occur indoors, signal loss and sig-229
nal noise are prevalent in human trajectory data (Hwang et al. 2018). Additionally,230
trajectory data recorded with smart phone applications are also influenced by other231
factors (e.g. battery capacity of smart phones). Uncertainties in the individuals’ tra-232
jectories make it complicated to explore their travel mode choice behavior patterns.233
Therefore, the features are interpolated for the days of data loss. We propose a solu-234
tion to interpolate the gaps based on the previous and next workdays (or weekends).235
The assumption is that human mobility patterns are regular, periodic and predictable,236
which has been demonstrated by several related studies (Gonzalez et al. 2008; Song237
et al. 2010; Yuan and Raubal 2016). Considering that people’s travel habits are238
normally stable, we choose the features of the adjacent days to impute the gaps.239
Concretely, the gap will be interpolated with the features of two workdays adjacent240
to it if the trajectories are missing on a workday. Likewise, the data loss on weekends241
is processed in a similar way.242
4.3 Measuring Individuals’ Similarity Based on243
Autocorrelation244
Based on the aforementioned features, the goal of this section is to measure the245
similarity between different users. Since each feature is in the form of a time series, we246
calculate the similarities of the time series based on their autocorrelation coefficients247
(AC) (D’Urso and Maharaj 2009). Compared to similarity measurement methods248
based on the shape of the time series, autocorrelation considers interior correlation249
characteristics of the time series. Given a set of time series X={x1,x2,...,xK},250
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
8P. Zhao et al.
(a) duration
(b) distance
Fig. 2 Temporal variations of the duration and distance by means of various travel modes for two
exemplary users (red and blue)
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 9
as shown in Eq. (3), the autocorrelation coefficient of the kth time series at time lag251
rcan be expressed as Eq. (4):252
X=
x11 ... xk1... xK1
.
.
..
.
..
.
.
x1t... xkt ... xKt
.
.
..
.
..
.
.
x1T... xkT ... xKT
(3)
253
ˆρkr =T
t=r+1(xkt xk)(xk(tr)xk)
T
t=1(xkt xk)2(4)254
where Kis the number of time series, Tis the length of one time series, xk={xkt :255
t=1,2,...,T}represents the kth (k=1,2,...,K)time series, xkt stands for the256
tth observation value of the kth time series, and xkis the mean of the kth time series.257
As we want to classify people not solely on the regularity of their transport mode258
choices, but also by taking into account the relative number of times they used a259
certain transport mode, we introduce a user-specific weighting of each autocorre-260
lation vector ˆρk. Namely, as each autocorrelation vector is computed from either261
the daily distance or duration of a single mode of transport, we multiply it by the262
share of this mode’s distance or duration over the whole study period for each user263
(wk=dk/10
k=1dk, where dkis either the distance or duration of transport mode k,264
depending on how the respective autocorrelation vector was computed). To clarify,265
let us assume there are two users: one never uses the car, while the other uses the car266
every day to go to work. Simply looking at the autocorrelation would classify these267
users into the same class, as their time series are perfectly autocorrelated. Introducing268
user-specific weights for the autocorrelation, essentially scales the autocorrelation269
values of the first user to zero, thus increasing the likelihood that this user gets into270
another cluster.271
Given two sets of time series for two users’ travel duration by use of various travel272
modes and the determined time lag R, two sets of autocorrelation vectors can be273
calculated accordingly on the basis of Eq. (4). We assume the autocorrelation vectors274
ˆ
ρsi =(ˆρsi1,..., ˆρsi R)and ˆ
ρti =(ˆρti1,..., ˆρtiR )correspond to the ith travel mode275
for users sand trespectively, and the weights of them are ws=(ws1,...,w
s10)276
and wt=(wt1,...,w
t10). The similarity of two users is measured based on the277
autocorrelation vectors and weights, which is denoted by the follow formula:278
d2
st =
10
i=1
R
r=1
(ˆρsir ·wsi −ˆρtir ·wti)2(5)279
where i=1,2,...,10 stands for the ith travel mode, r=1,2,...,Rare the time280
lags for the time series.281
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
10 P. Zhao et al.
4.4 Clustering the Time Series282
We measure the similarities of individuals by calculating the distance of their cor-283
responding weighted autocorrelation vectors at different time lags with Eq.5and284
obtain the similarity matrix. On the basis of the similarity matrix, the individuals285
can be categorized into different groups based on their travel mode choice behavior286
changes. In this work, we choose a hierarchical clustering algorithm to process the287
similarity matrix (Gower and Ross 1969). Hierarchical clustering algorithms attempt288
to construct a hierarchy of clusters through a bottom-up (i.e. agglomerative) or top-289
down strategy (i.e. divisive). The output of the algorithm is a dendrogram, which290
intuitively displays the hierarchical relationships between the clusters. One of its291
major advantages is that hierarchical clustering is flexible for inputting a similarity292
matrix generated by various similarity or distance measures according to the research293
purpose. Therefore, it has been widely used in human mobility analysis (Shen and294
Cheng 2016; Wang et al. 2018).295
The number of clusters to be generated in this study is determined by the Calinski-296
Harabasz Index (CH-index) (Cali´nski and Harabasz 1974), which evaluates how297
well the dataset is separated quantitatively. Normally, the location of the maximum298
CH-index corresponds to the optimal number of clusters. Before calculating the299
CH-index, two basic elements, namely SSW and SSB (sum of squares within resp.300
between the clusters) need to be calculated, which can be used to quantify the overall301
within-cluster and between-cluster variances. The CH-index is a ratio based on them.302
Given a set X={x1,x2,...,xN}, representing a dataset with Ndata points, X=303
N
i=1xi/Nis the center of the whole dataset. Let us denote the centroids of clusters304
{C1,C2,...,CM}as C={c1,c2,...,cM}, with Mbeing the number of clusters and305
nibeing the number of elements in cluster Ci.SSW,SSB and CH are then computed306
as follows:307
SSW =
N
i=1
xiCpi
2(6)308
SSB =
M
i=1
ni
ciX
2(7)309
CH =SSB/(M1)
SSW/(NM)(8)
310
where xirepresents the ith point, Cpiindicates the centroid of the pth cluster and xi
311
is in the pth cluster.312
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 11
5 Results313
5.1 Measuring the Similarity of Individuals314
This section aims at measuring the individuals’ similarity with autocorrelation based315
on the constructed time series that represent their travel mode choice behavior, which316
will provide input for the clustering analysis. The autocorrelation coefficients at dif-317
ferent time lags constitute the autocorrelation vector of a time series. Theoretically,318
the time lag ris between 1 and T1 (where Tis the number of days in the time319
series). Considering the specific characteristics of the constructed features (e.g. reg-320
ularity and periodicity), we examine the differences of their autocorrelation coeffi-321
cients at different time lag limits. We select all users’ travel duration and distance by322
train as an example. The variations of autocorrelation coefficients for duration and323
distance at different time lags are shown in Fig.3. To observe the variations of the324
autocorrelation coefficients more clearly, Fig. 3(c) and (d) visualize the first 50 auto-325
correlation coefficients. It can be seen that the autocorrelation coefficients display326
regular weekly patterns. According to the recommendation by (Box et al. 2015) that327
the number of lags up to about a quarter of the time series length is sufficient to assess328
the dependence structure of the time series, this study examine the autocorrelation329
coefficients up to lags 7, 14, 21, and 28 for the calculations of similarity.330
Based on the selected maximum time lag, similarity matrices for both duration and331
distance can be calculated using the procedure introduced in Sect. 4.3. Here, the two332
similarity matrices are calculated based on the autocorrelation coefficients up to lags333
28 to reflect the overall similarity of individuals’ travel mode choice behavior in terms334
of duration and distance. In addition, we further calculate the correlation coefficient335
of the two matrices. The correlation coefficient reaches 0.86, which implies that travel336
duration and distance are comparatively consistent in the depiction of individuals’337
travel mode choice behavior.338
5.2 Detecting Travel Mode Choice Behavior Patterns339
On the basis of similarity matrices of duration-based and distance-based features,340
the goal of this section is to segregate all individuals into groups. The individuals are341
divided into subgroups in the form of a hierarchical tree. The hierarchical tree can be342
cut at certain predetermined locations to divide the whole dataset into several groups.343
We utilize the CH-index to determine the number of clusters, which evaluates the344
clustering validity based on the average between- and within- cluster sum of squares.345
Based on the similarity matrices, Fig. 4presents the relations between the number346
of clusters and CH-index for different time lags. Note that the optimal number of347
clusters is different for the different number of time lags. According to the optimal348
number of clusters determined by Fig. 4, Table 1displays the clustering results for349
different time lags. Specifically, the clusters with one or two users are regarded as350
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
12 P. Zhao et al.
(a) Autocorrelation vectors of travel duration (b) Autocorrelation vectors of travel distance
(c) The first 50 autocorrelation coefficients (d) The first 50 autocorrelation coefficients
Fig. 3 Variations of the autocorrelation coefficients at different time lags for all users
outliers. Next, we analyze the users’ travel mode choice behavior patterns at the351
aggregate and individual levels respectively based on the fourth clustering results352
(i.e. the number of lags is 28).353
First, we analyze the travel mode choice patterns at the aggregate level. Specifi-354
cally, the proportions of travel duration and distance by means of various travel modes355
are calculated for each cluster, as shown in Table 2. It can be observed that the indi-356
viduals in each cluster show different transport mode use patterns. For example, car357
and train are chosen as the main travel modes, and e-car and walk as secondary travel358
modes in cluster 1. For cluster 2, train occupies a high proportion as the main travel359
mode, while car, e-car and walk occupy comparatively low proportions as secondary360
travel modes. Meanwhile, note that the two clusters based on duration show the same361
main travel modes as those based on distance, namely car and train, train, as well as362
car. It also demonstrates that travel duration and distance are consistent in describing363
individuals’ travel mode choice behavior. However, the difference between them is364
also worth noting. For instance, walk does not appear as secondary travel mode in365
the clusters based on distance, which is also in accordance with individuals’ daily366
trips. After all, walk is normally selected for short-distance trips.367
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 13
0 4 8 12 16 20 24 28 30
The number of clusters
4
6
8
10
12
14
CH-index
Lags 7
Lags 14
Lags 21
Lags 28
(a) duration
0 4 8 12 16 20 24 28 30
The number of clusters
4
5
6
7
8
9
10
11
CH-index
Lags 7
Lags 14
Lags 21
Lags 28
(b) distance
Fig. 4 Relation graph of the number of clusters and CH-index
Tabl e 1 Results of the clustering for different time lags
Num(lags) = 7 Num(lags) = 14 Num(lags) = 21 Num(lags) = 28
Cluster Num Cluster Num Cluster Num Cluster Num
Duration 1100 1100 1100 1100
2 6 2 5 2 5 2 5
3 1 3 1 3 1 3 1
4 1 4 1 4 1
Distance 1104 1104 194 193
2 2 2 2 2 9 2 9
3 1 3 1 3 2 3 1
4 1 4 2
5 1 5 1
6 1
Second, travel mode choice behavior patterns at the individual level are investi-368
gated. Concretely, five human mobility indicators are selected to explore individuals’369
travel mode choice behavior patterns, including travel duration and distance based370
on four travel modes (i.e. car, e-car, train and walk) as well as carbon dioxide (CO2)371
emissions. These emissions were computed by multiplying the distances covered372
with each means of transport by a mode-specific constant as given by the Swiss plat-373
form for mobility management tools (mobitool) (Tuchschmid and Halder 2010). The374
nine resulting indicators refer to the mean of the corresponding observation values375
in the selected period for each user. In addition, we normalize these indicators to376
ensure that they fall between 0 and 1 in order to conveniently compare them in the377
same figure. To understand the distribution of indicators for the users in each cluster,378
we visualize their distribution using a boxplot in Fig. 5. Looking at Fig. 5a, it can379
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
14 P. Zhao et al.
Tabl e 2 The shares of travel modes for different clusters
Duration Distance
Travel mode Cluster 1 Cluster 2 Cluster 1 Cluster 2
Bicycle 0.038 0.021 0.017 0.008
Boat 0.009 0.015 0.003 0.001
Bus 0.014 0.015 0.008 0.005
Car 0.314 0.146 0.373 0.208
Coach 0.003 0.005 0.004 0.000
e-bicycle 0.000 0.000 0.000 0.000
e-car 0.173 0.129 0.210 0.131
Train 0.252 0.488 0.351 0.622
Tram 0.008 0.008 0.006 0.003
Wal k 0.189 0.173 0.028 0.022
(a) duration (b) distance
Fig. 5 Boxplots of five mobility indicators in different clusters
be seen that the users in clusters 1 spend more time by car for their trips than those380
of cluster 2 as a whole. Similarly, we can conclude that the users in cluster 2 spend381
more time by train for their trips than those of clusters 1, which results from train as382
the sole main travel mode for the users in cluster 2. It is noteworthy that the travel383
time by e-car and walk is not very different for the users in the two clusters, which384
is due to the fact that e-car and walk are both regarded as secondary travel modes.385
Additionally, another remarkable phenomenon is seen in the comparison of CO2
386
emissions among the two clusters. High CO2emissions of clusters 1 indicate that387
car is probably still the main source of transport-related carbon dioxide emissions.388
Public transport and green travel modes (e.g. e-car) are efficient alternatives for the
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 15
reduction of CO2emissions and for sustainable urban development. Similar anal-389
yses can be conducted to explain the travel mode choice behavior patterns for the390
distance-based indicators.391
6 Conclusions and Future Work392
The availability of long-term human movement data with semantic information (e.g.393
travel mode) enables us to investigate individuals’ travel mode choice behavior pat-394
terns. Specifically, behavioral change processes cannot be considered in traditional395
travel mode choice studies due to a lack of advanced data collection methods. In this396
study, we propose a clustering-based framework to understand individuals’ travel397
mode choice behavior patterns by segregating them into groups that exhibit simi-398
lar travel mode choice behavior, based on a human trajectory dataset with semantic399
information. First, we construct duration-based and distance-based features in the400
form of time series to depict individuals’ travel mode choice behavior. Next, we401
propose a weighted autocorrelation method to measure the similarity of individuals.402
Finally, a hierarchical clustering algorithm is employed to divide the individuals into403
groups based on the similarity matrix. For a case study in Switzerland, three clusters404
of individuals are detected and interpreted from the perspective of travel mode choice405
behavior patterns at the aggregate and individual levels. Our contributions facilitate406
understanding people’s travel mode choice behavior in multimodal transportation407
and how to design reasonable alternatives to private cars for more sustainable cities.408
In addition, dividing the individuals into different groups based on their travel mode409
choice behavior will also help policymakers design and provide personalized travel410
mode recommendation services for different user groups.411
However, there are several limitations in the current study, which could be regarded412
as directions for future work. First, the interpolation of data loss is simply conducted413
based on the historical movement data. Especially, only the data of two adjacent414
days are used, the effect of which requires to be considered. Hence, new methods415
and techniques of data gaps imputation should be investigated to impute gaps in416
the trajectory dataset. Second, the current hierarchical clustering algorithm has lim-417
ited ability working on high-dimension datasets. New advanced high-dimensional418
clustering methods would be more appropriate for this study. Last but not least, this419
study only analyzes the individuals’ travel mode choice behavior patterns based on420
the clustering results. Although it can reflect the difference between individuals, it421
would be more meaningful to explore the cross influence of socio-demographic and422
urban environment factors on the patterns.423
Acknowledgements This research was supported by the Swiss Data Science Center (SDSC), by424
the Swiss Innovation Agency Innosuisse within the Swiss Competence Center for Energy Research425
(SCCER) Mobility and by the Swiss Federal Railways SBB.426
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
16 P. Zhao et al.
References427
Aghabozorgi S, Shirkhorshidi AS, Wah TY (2015) Time-series clustering-a decade review. Inf Syst428
53:16–38429
An S, WangZ, Cui J (2015) Integrating regret psychology to travel mode choice for a transit-oriented430
evacuation strategy. Sustainability 7(7):8116–8131431
Bankó Z, Abonyi J (2012) Correlation based dynamic time warping of multivariate time series.432
Expert Syst Appl 39(17):12814–12823433
Barbosa H, Barthelemy M, Ghoshal G, James CR, Lenormand M, Louail T, Menezes R, Ramasco434
JJ, Simini F, Tomasini M (2018) Human mobility: models and applications. Phys Rep435
Barragan JF, Fontes CH, Embiruçu M (2016) A wavelet-based clustering of multivariate time series436
using a multiscale spca approach. Comput Ind Eng 95:144–155437
Böcker L, Prillwitz J, Dijst M (2013) Climate change impacts on mode choices and travelled438
distances: a comparison of present with 2050 weather conditions for the randstad holland. J439
Transp Geogr 28:176–185440
Böcker L, van Amen P, Helbich M (2017) Elderly travel frequencies and transport mode choices in441
Greater Rotterdam, the Netherlands. Transportation 44(4):831–852442
Box GE, Jenkins GM, Reinsel GC, Ljung GM (2015) Time series analysis: forecasting and control.443
John Wiley, New Jersey444
Bucher D, Cellina F, Mangili F, Raubal M, Rudel R, Rizzoli AE, Elabed O (2016) Exploiting445
fitness apps for sustainable mobility-challenges deploying the Goeco! app. ICT for sustainability446
(ICT4S)447
Bucher D, Mangili F, Cellina F, Bonesana C, Jonietz D, Raubal M (2019) From location tracking to448
personalized eco-feedback: a framework for geographic information collection, processing and449
visualization to promote sustainable mobility behaviors. Travel Behav Soc 14:43–56450
Cali´nski T, Harabasz J (1974) A dendrite method for cluster analysis. Commun Stat-Theory Methods451
3(1):1–27452
Chen C, Gong H, Paaswell R (2008) Role of the built environment on mode choice decisions:453
additional evidence on the impact of density. Transportation 35(3):285–299454
Daisy NS, Millward H, Liu L (2018) Trip chaining and tour mode choice of non-workers grouped455
by daily activity patterns. J Transp Geogr 69:150–162456
Dias JG, Vermunt JK, Ramos S (2015) Clustering financial time series: new insights from an457
extended hidden markov model. Eur J Oper Res 243(3):852–864458
Ding C, Wang D, Liu C, Zhang Y, Yang J (2017) Exploring the influence of built environment on459
travel mode choice considering the mediating effects of car ownership and travel distance. Transp460
Res Part A: Policy Pract 100:65–80461
Ding L, Zhang N (2016) A travel mode choice model using individual grouping based on cluster462
analysis. Procedia Eng 137:786–795463
D’Urso P, Maharaj EA (2009) Autocorrelation-based fuzzy clustering of time series. Fuzzy Sets464
Syst 160(24):3565–3589465
Froehlich J, Dillahunt T, Klasnja P, Mankoff J, Consolvo S, Harrison B, Landay JA (2009) Ubigreen:466
investigatinga mobile tool for tracking and supporting green transportation habits. In: Proceedings467
of the SIGCHI conference on human factors in computing systems, ACM, pp 1043–1052468
Gao Z-K, Yang Y-X, Fang P-C, Jin N-D, Xia C-Y, Hu L-D (2015) Multi-frequency complex network469
from time series for uncovering oil-water flow structure. Sci Rep 5:8222470
Glasser W (1999) Choice theory: a new psychology of personal freedom. Harper Perennial, New471
York472
Gonzalez MC, Hidalgo CA, Barabasi A-L (2008) Understanding individual human mobility pat-473
terns. Nature 453(7196):779474
Górecki T (2018) Classification of time series using combination of DTW and LCSS dissimilarity475
measures. Commun Stat-Simul Comput 47(1):263–276476
Gower JC, Ross GJ (1969) Minimum spanning trees and single linkage cluster analysis. Appl Stat477
54–64478
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
A Clustering-Based Framework for Understanding Individuals’ Travel Mode … 17
Gunopulos D, Das G (2001) Time series similarity measures and time series indexing. Acm Sigmod
479
Record, vol 30, ACM, p 624480
Han Y, Li W, Wei S, Zhang T (2018) Research on passenger’s travel mode choice behavior waiting481
at bus station based on sem-logit integration model. Sustainability 10(6):1996482
Heinen E, Chatterjee K (2015) The same mode again? an exploration of mode choice variability in483
great britain using the national travel survey. Transp Res Part A: Policy Pract 78:266–282484
Huang H, Gartner G, Krisp JM, Raubal M, de Weghe NV (2018) Location based services: ongoing485
evolution and research agenda. J Locat Based Serv 12(2):63–93486
Hunecke M, Blöbaum A, Matthies E, Höger R (2001) Responsibility and environment: ecological487
norm orientation and external factors in the domain of travel mode choice behavior. Environ488
Behav 33(6):830–852489
Hwang S, VanDeMark C, Dhatt N, Yalla SV, Crews RT (2018) Segmenting human trajectory data490
by movement states while addressing signal loss and signal noise. Int J Geogr Inf Sci 32(7):1391–491
1412492
Jonietz D, Bucher D (2018) Continuous trajectory pattern mining for mobility behaviour change493
detection. In: LBS 2018: 14th international conference on location based services. Springer, pp494
211–230495
Jonietz D, Bucher D, Martin H, Raubal M (2018) Identifying and interpreting clusters of persons496
with similar mobility behaviour change processes. In: Mansourian A, Pilesjö P, Harrie L, van497
Lammeren R (eds) AGILE 2018—geospatial technologies for all. Springer International Pub-498
lishing, Cham, pp 291–307499
Klinger T (2017) Moving from monomodality to multimodality? changes in mode choice of new500
residents. Transp Res Part A: Policy Pract 104:221–237501
Liang Q, Weng J, Zhou W, Santamaria SB, Ma J, Rong J (2018) Individual travel behavior modeling502
of public transport passenger based on graph construction. J Adv Transp 2018503
Liu C, Susilo YO, Karlström A (2015) The influence of weather characteristics variability on504
individual’stravel mode choice in different seasons and regions in Sweden. Transp Policy 41:147–505
158506
Łuczak M (2016) Hierarchical clustering of time series data with parametric derivative dynamic507
time warping. Expert Syst Appl 62:116–130508
Murtagh N, Gatersleben B, Uzzell D (2012) Multiple identities and travel mode choice for regular509
journeys. Transp Res Part F: Traffic Psychol Behav 15(5):514–524510
Shen J, Cheng T (2016) A framework for identifying activity groups from individual space-time511
profiles. Int J Geogr Inf Sci 30(9):1785–1805512
Siła-Nowicka K, Vandrol J, Oshan T, Long JA, Demšar U, Fotheringham AS (2016) Analysis of513
human mobility patterns from gps trajectories and contextual information. Int J Geogr Inf Sci514
30(5):881–906515
Song C, Qu Z, Blumm N, Barabási A-L (2010) Limits of predictability in human mobility. Science516
327(5968):1018–1021517
Sun B, Ermagun A, Dan B (2017) Built environmental impacts on commuting mode choice and518
distance: evidence from Shanghai. Transp Res Part D: Transp Environ 52:441–453519
Tuchschmid M, Halder M (2010) mobitool–grundlagenbericht: Hintergrund. Methodik & Emis-520
sionsfaktoren521
Vij A, Carrel A, Walker JL (2013) Incorporating the influence of latent modal preferences on travel522
mode choice behavior. Transp Res Part A: Policy Pract 54:164–178523
Wang Y, Qin K, Chen Y, Zhao P (2018) Detecting anomalous trajectories and behavior patterns524
using hierarchical clustering from taxi GPS data. ISPRS Int J Geo-Inf 7(1):25525
Weiser P, Scheider S, Bucher D, Kiefer P, Raubal M (2016) Towards sustainable mobility behavior:526
research challenges for location-aware information and communication technology. GeoInfor-527
matica 20(2):213–239528
Xiong Y, Yeung D-Y (2002) Mixtures of arma models for model-based time series clustering. In:529
2002 IEEE International Conference on Data Mining, 2002. ICDM 2003. Proceedings, IEEE, pp530
717–720531
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
UNCORRECTED PROOF
18 P. Zhao et al.
Ye Y, Niu C, Jiang J, Ge B, Yang K (2017) A shape based similarity measure for time series
532
classification with weighted dynamic time warping algorithm. In: 4th International conference533
on information science and control engineering (ICISCE), 2017, IEEE, pp 104–109534
Yuan Y, Raubal M (2012) A framework for spatio-temporal clustering from mobile phone data.535
Workshop on complex data mining in a geospatial context proceedings at AGILE 2012. Associ-536
ation of Geographic Information Laboratories for Europe (AGILE), pp 22–26537
Yuan Y,Raubal M (2014) Measuring similarity of mobile phone user trajectories—a spatio-temporal538
edit distance method. Int J Geogr Inf Sci 28(3):496–520539
Yuan Y, Raubal M (2016) Analyzing the distribution of human activity space from mobile phone540
usage: an individual and urban-oriented study. Int J Geogr Inf Sci 30(8):1594–1621541
Yue M, Kang C, Andris C, Qin K, Liu Y, Meng Q (2018) Understanding the interplay between bus,542
metro, and cab ridership dynamics in shenzhen, China. Trans GIS 22(3):855–871543
Zhao P, Kwan M-P, Qin K (2017a) Uncovering the spatiotemporal patterns of CO2 emissions by544
taxis based on individuals’ daily travel. J Transp Geogr 62:122–135545
Zhao P, Qin K, Ye X, Wang Y, Chen Y (2017b) A trajectory clustering approach based on decision546
graph and data field for detecting hotspots. Int J Geogr Inf Sci 31(6):1101–1127547
475797_1_En_5_Chapter TYPESET DISK LE CP Disp.:23/3/2019 Pages: 18 Layout: T1-Standard
Author Proof
... Street parking availability has proven itself to be a critical part of an urban transportation system [25]. Their creation was similar to another one already created but with improved features [26]. Regardless of how fast businesses spread, the central business districts still and always will attract numerous customers who had little to no choice but to park on the streets. ...
Article
Over time, scientists and engineers have developed technology that has improved everyday living. For every invention, they have found to reinvent it and create a better version of it. An example would be parking lots. From the usual time-in-time-out system; window cashier to parking spot back to window cashier, the hassle of this cycle was reduced through the use of license plate recognition which this study tackled. License plate identification technology has become a useful factor in different issues like parking management, vehicle owner recognition, etc. Computer vision has provided a means to reduce manual labor in parking lots by replacing them with sensors and scanners which would take over. This would be beneficial to the company as this technology would require minimal maintenance. Through image processing and enhancement, the license plates of these cars could be taken account of in real-time and this data could be used to generate the parking bills given the time-in and a time-out of the vehicle. This paper aimed to tackle the problems of car management and to further simplify the parking system. © 2020, World Academy of Research in Science and Engineering. All rights reserved.
... An anomalous deviation of these features from one week to another can indicate a transition from one phase of behavior change to another and should be reflected within the supporting ICT. Additionally, identifying people in similar behavioral transition phases can be used for analytical purposes or to target individual groups with specific incentives (Zhao et al. 2019). ...
Chapter
Full-text available
Urban mobility and the transport of people have been increasing in volume inexorably for decades. Despite the advantages and opportunities mobility has brought to our society, there are also severe drawbacks such as the transport sector’s role as one of the main contributors to greenhouse-gas emissions and traffic jams. In the future, an increasing number of people will be living in large urban settings, and therefore, these problems must be solved to assure livable environments. The rapid progress of information and communication, and geographic information technologies, has paved the way for urban informatics and smart cities, which allow for large-scale urban analytics as well as supporting people in their complex mobile decision making. This chapter demonstrates how geosmartness, a combination of novel spatial-data sources, computational methods, and geospatial technologies, provides opportunities for scientists to perform large-scale spatio-temporal analyses of mobility patterns as well as to investigate people’s mobile decision making. Mobility-pattern analysis is necessary for evaluating real-time situations and for making predictions regarding future states. These analyses can also help detect behavioral changes, such as the impact of people’s travel habits or novel travel options, possibly leading to more sustainable forms of transport. Mobile technologies provide novel ways of user support. Examples cover movement-data analysis within the context of multi-modal and energy-efficient mobility, as well as mobile decision-making support through gaze-based interaction.
Article
Full-text available
The elderly population is increasing rapidly. Understanding travel behaviour for this group of commuters (in terms of the trip purpose and travel time) is necessary for future transport planning. Many researchers are working on travel’s spatial and temporal analysis to provide operational decision making and transport network planning. This research study’s primary purpose is to identify the influence of trip duration (using public transport), time of the day (usage of public transport), and public transport (PT) accessibility over public transport mode preference by elderly (over 65 years of age) commuters. The methodology of this study is divided into two parts as spatial analysis and temporal analysis. The research identified the dependency of trip duration, time of the day, geographical areas, and PT access over transport mode preference of elderly. The temporal study shows that transport mode preference can vary depending on trip purposes. However, for specific trip durations and times of the day, the elderly sometimes choose PT as a mobility mode. For instance, on shopping trips between 10:00 and 11:00 a.m., the elderly have a greater possibility of choosing public transport over private vehicles. Moreover, the results show the public transport mode preference based on different times of the day and trip purposes. Urban and transport planner can use the results to modify/plan public transport schedule, which can be easily accessible by the elderly population.
Article
Full-text available
Nowadays, most people carry around a powerful smartphone which is well suited to constantly monitor the location and sometimes even the activity of its user. This makes tracking prevalent and leads to a large number of projects concerned with trajectory data. One area of particular interest is transport and mobility, where data is important for urban planning and smart city-related activities, but can also be used to provide individual users with feedback and suggestions for personal behavior change. As part of a large-scale study based in Switzerland, we use activity tracking data to provide people with eco-feedback on their own mobility patterns and stimulate them to adopt more energy-efficient mobility choices. In this paper we explore the opportunities offered by smartphone based activity tracking, propose a general framework to exploit location data to foster more sustainable mobility behavior, describe the technical solutions chosen and discuss a range of outcomes in terms of user perception and sustainability potential. The presented approach extracts mobility patterns from users’ trajectories, computes credible alternative transport options, and presents the results in a concise and clear way. The resulting eco-feedback helps people to understand their mobility choices, discover the most non-ecological parts of their travel behavior, and explore feasible alternatives.
Article
Full-text available
We are now living in a mobile information era, which is fundamentally changing science and society. Location Based Services (LBS), which deliver information depending on the location of the (mobile) device and user, play a key role in this mobile information era. This article first reviews the ongoing evolution and research trends of the scientific field of LBS in the past years. To motivate further LBS research and stimulate collective efforts, this article then presents a series of key research challenges that are essential to advance the development of LBS, setting a research agenda for LBS to ‘positively’ shape the future of our mobile information society. These research challenges cover issues related to the core of LBS development (e.g. positioning, modelling, and communication), evaluation, and analysis of LBS-generated data, as well as social, ethical, and behavioural issues that rise as LBS enter into people’s daily lives.
Article
Full-text available
To improve the mode share of public transport and reduce the transition to private transport of passengers waiting at bus station, the mechanism of passengers' decision-making procedure and influence factors of the travel mode choice were analyzed. Some latent variables such as safety, comfort, convenience, flexibility and economy were selected to reflect the satisfaction degree of passengers on the service level of public transport. Taking Jinan City as an example, the questionnaire of passengers' travel choice behavior at bus station was designed and carried out. Based on the structure equation model (SEM), the relationship between the satisfaction degree and some latent variables such as safety and comfort was discussed. The SEM method analysis shows that, of the influence level of the latent variables to the service level of public transport, flexibility is the most significant variable affecting passenger's satisfaction degree followed by safety, convenience, comfort and economy. Travel mode choice model of passengers waiting at bus station was established with an integration approach of SEM and nested logit (NL) model. The SEM-NL integration model results reveal that gender, monthly income, purpose of the trip, travel distance, safety and convenience service level have a significant effect on the choice of the upper model (public transport or private transport). Passenger's age, vehicle ownership and bus ride frequency have great influence on the choice of the lower mode (ORB: original route bus; ARB: alternative route bus; Taxi; and Shared bike). Sensitivity analysis reveals that the transition probabilities from private transport to public transport can reach the highest point (respectively, 69.85%, 68.84% and 35.51%) when safety service reaches level 4, convenience service reaches level 3, or comfort service reaches level 2, indicating that the safety level equal to 4, convenience level equal to 3 and comfort level equal to 2 are the key threshold to increase the public transport mode share. Some proposals such as ensuring good accessibility of public transport, shortening the transfer distance of different routes, creating a comfortable travel environment and integrating bus ticket system have been put forward for the sustainable development of public transport system.
Article
Full-text available
This paper presents a novel method for mining the individual travel behavior regularity of different public transport passengers through constructing travel behavior graph based model. The individual travel behavior graph is developed to represent spatial positions, time distributions, and travel routes and further forecasts the public transport passenger’s behavior choice. The proposed travel behavior graph is composed of macronodes, arcs, and transfer probability. Each macronode corresponds to a travel association map and represents a travel behavior. A travel association map also contains its own nodes. The nodes of a travel association map are created when the processed travel chain data shows significant change. Thus, each node of three layers represents a significant change of spatial travel positions, travel time, and routes, respectively. Since a travel association map represents a travel behavior, the graph can be considered a sequence of travel behaviors. Through integrating travel association map and calculating the probabilities of the arcs, it is possible to construct a unique travel behavior graph for each passenger. The data used in this study are multimode data matched by certain rules based on the data of public transport smart card transactions and network features. The case study results show that graph based method to model the individual travel behavior of public transport passengers is effective and feasible. Travel behavior graphs support customized public transport travel characteristics analysis and demand prediction.
Chapter
Full-text available
With the emergence of new mobility options and various initiatives to increase the sustainability of our travel behaviour, it is desirable to gain a deeper understanding of our behavioural reactions to such stimuli. Although it is now possible to use GPS-tracking to record people’s movement behaviour over a longer period, there is still a lack of computational methods which allow to detect and evaluate such behaviour change processes in the resulting datasets. In this study, we propose a data mining method for describing individual persons’ mobility behaviour change processes based on their movement trajectories and clustering participants based on the similarity of these behavioural adaptations. We further propose to use a decision tree classifier to semantically explain the derived clusters in a human-interpretable form. We apply our method to a real, longitudinal movement dataset.
Article
Full-text available
Anomalous taxi trajectories are those chosen by a small number of drivers that are different from the regular choices of other drivers. These anomalous driving trajectories provide us an opportunity to extract driver or passenger behaviors and monitor adverse urban traffic events. Because various trajectory clustering methods have previously proven to be an effective means to analyze similarities and anomalies within taxi GPS trajectory data, we focus on the problem of detecting anomalous taxi trajectories, and we develop our trajectory clustering method based on the edit distance and hierarchical clustering. To achieve this objective, first, we obtain all the taxi trajectories crossing the same source–destination pairs from taxi trajectories and take these trajectories as clustering objects. Second, an edit distance algorithm is modified to measure the similarity of the trajectories. Then, we distinguish regular trajectories and anomalous trajectories by applying adaptive hierarchical clustering based on an optimal number of clusters. Moreover, we further analyze these anomalous trajectories and discover four anomalous behavior patterns to speculate on the cause of an anomaly based on statistical indicators of time and length. The experimental results show that the proposed method can effectively detect anomalous trajectories and can be used to infer clearly fraudulent driving routes and the occurrence of adverse traffic events.
Article
Full-text available
This paper considers the problem of partitioning an individual GPS trajectory data into homogeneous, meaningful segments such as stops and trips. Signal loss and signal noise are highly prevalent in human trajectory data, and it is challenging to deal with uncertainties in segmentation algorithms. We propose a new trajectory segmentation algorithm that detects stop segments in a noise-robust manner from GPS data with time gaps. The algorithm consists of three steps that impute time gaps, split data into base segments and estimate states over a base segment. The state-dependent path interpolation was proposed as a framework for gap imputation to deal with locational and temporal uncertainties associated with signal loss. A spatiotemporal clustering-based tra-jectory segmentation was proposed to detect spatiotemporal clusters of any shape regardless of density to cut a trajectory into internally similar base segments. Fuzzy inference was employed to deal with borderline cases in determining states over base segments based on input features. The proposed algorithm was applied to detect stop/move episodes from raw GPS trajectories that were collected from 20 urban and 19 suburban participants. Sensitivity analysis was conducted to guide the choice of parameters such as the temporal and spatial definitions of a stop. Experimentation results show that the proposed method correctly identified 92% of stop/move episodes, and correctly estimated 98% of episode duration. This study indicates that a sequence of state-dependent gap imputation, clustering-based data segmentation and fuzzy-set-based state estimation can satisfactorily deal with uncertainty in processing human GPS trajectory data. ARTICLE HISTORY
Chapter
Full-text available
With the emergence of ubiquitous movement tracking technologies, developing systems which continuously monitor or even influence the mobility behaviour of individuals in order to increase its sustainability is now possible. Currently, however, most approaches do not move beyond merely describing the status quo of the observed mobility behaviour, and require an expert to assess possible behaviour changes of individual persons. Especially today, automated methods for this assessment are needed, which is why we propose a framework for detecting behavioural anomalies of individual users by continuously mining their movement trajectory data streams. For this, a workflow is presented which integrates data preprocessing, completeness assessment, feature extraction and pattern mining, and anomaly detection. In order to demonstrate its functionality and practical value, we apply our system to a real-world, large-scale trajectory dataset collected from 139 users over 3 months.
Article
The most common mass transit modes in metropolitan cities include buses, subways, and taxicabs, each of which contribute to an interconnected complex network that delivers urban dwellers to their destinations. Understanding the intertwined usages of these three transit modes at different places and time allows for better sensing of urban mobility and the built environment. In this article, we leverage a comprehensive data collection of bus, metro, and taxicab ridership from Shenzhen, China to unveil the spatio‐temporal interplay between different mass transit modes. To achieve this goal, we develop a novel spectral clustering framework that imposes spatio‐temporal similarities between mass transit mode usage in urban space and differentiates urban spaces associated with distinct ridership patterns of mass transit modes. Five resulting categories of urban spaces are identified and interpreted with auxiliary knowledge of the city's metro network and land‐use functionality. In general, different categorized urban spaces are associated with different accessibility levels (such as high‐, medium‐, and low‐ranked) and different urban functionalities (such as residential, commercial, leisure‐dominant, and home–work balanced). The results indicate that different mass transit modes cooperate or compete based on demographic and socioeconomic attributes of the underlying urban environments. Our proposed analytical framework provides a novel and effective way to explore the mass transit system and the functional heterogeneity in cities. It demonstrates great potential for assisting policymakers and municipal managers in optimizing public transportation facility allocation and city‐wide daily commuting distribution.
Article
Transportation planners are increasingly looking to reduce drive-alone travel. However, the existence of complex tours could be a major barrier to the shift from drive-alone to public transport. This is particularly problematical for the non-worker segment of the population (most of whom are women and/or older adults). To better understand how socio-demographics, trip attributes, and land use pattern shape tour complexity and mode choices for individuals, this paper introduces a cluster-based approach to model tour complexity, trip chaining, and tour mode choice. Based on daily activity patterns and time-use, five non-worker clusters were identified from the Space-Time Activity Research (STAR) data for Halifax, Canada. Number of tours per day for all clusters was modeled using a Poisson regression model. Trip chaining was then modeled using an Ordered Probit model. Finally, tour mode choice was modeled using a Multinomial Logit (MNL) model. This framework utilizes home-based travel tours, the sequence of trips that begin and end at home, as the basic unit of analysis. Results show that socio-demographic characteristics and tour attributes are strong explanatory factors of travel behavior, consistent with existing literature. Also, urban form characteristics have significant influence on non-workers' travel behaviour and tour complexity. This paper also offers a typology of travel tours to account for different non-work travel purposes. The findings of this study will improve the future evaluation of transportation projects and land-use policymaking.