Conference PaperPDF Available

Vision Paper: Using Volunteered Geographic Information to Improve Mobility Prediction

  • AG

Abstract and Figures

Fine-grained real-time movement prediction is becoming increasingly important, with smartphones and vehicles constantly tracking our position and trying to guess our next location to timely provide us with recommendations, traffic forecasts, or driver assistance. Depending on the tracking accuracy, the recorded locations are first mapped to street segments, using a mobility model to choose the most likely road in case of ambiguities. The main prediction procedure uses a similar movement model (possibly incorporating additional user-specific data) to assess likely future travel choices. While the exact street topology is not essential on a very high level (e.g., when predicting the "next place" someone is going to be), it becomes more and more important if we try to predict the exact position of a person or vehicle. Similarly, different data sources (such as points of interest, land use zones, or building footprints) should be used for predictions at different levels of accuracy. In this paper, we assess current research trends concerning various types of volunteered geographical information (VGI), how this data can be used in different models to compute mobility predictions , and we present our vision for an integrated system that is able to use crowdsourced geographic data to perform mobility prediction at different levels.
Content may be subject to copyright.
Vision Paper: Using Volunteered Geographic
Information to Improve Mobility Prediction
Dominik Bucher
ETH Zurich, Institute of
Cartography and Geoinformation
Stefano-Franscini-Platz 5
Zurich, Switzerland 8093
Fine-grained real-time movement prediction is becoming increas-
ingly important, with smartphones and vehicles constantly track-
ing our position and trying to guess our next location to timely
provide us with recommendations, trac forecasts, or driver as-
sistance. Depending on the tracking accuracy, the recorded loca-
tions are rst mapped to street segments, using a mobility model
to choose the most likely road in case of ambiguities. The main pre-
diction procedure uses a similar movement model (possibly incor-
porating additional user-specic data) to assess likely future travel
choices. While the exact street topology is not essential on a very
high level (e.g., when predicting the “next place” someone is going
to be), it becomes more and more important if we try to predict
the exact position of a person or vehicle. Similarly, dierent data
sources (such as points of interest, land use zones, or building foot-
prints) should be used for predictions at dierent levels of accu-
racy. In this paper, we assess current research trends concerning
various types of volunteered geographical information (VGI), how
this data can be used in dierent models to compute mobility pre-
dictions, and we present our vision for an integrated system that
is able to use crowdsourced geographic data to perform mobility
prediction at dierent levels.
Applied computing Transportation;Information sys-
tems Location based services; Global positioning systems;
ACM proceedings, crowdsourcing, volunteered geographic infor-
mation, mobility, prediction
ACM Reference format:
Dominik Bucher. 2017. Vision Paper: Using Volunteered Geographic Infor-
mation to Improve Mobility Prediction. In Proceedings of PredictGIS’17:1st
ACM SIGSPATIAL Workshop on Prediction of Human Mobility , Redondo
Beach, CA, USA, November 7–10, 2017 (PredictGIS’17), 4 pages.
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full cita-
tion on the rst page. Copyrights for components of this work owned by others than
ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-
publish, to post on servers or to redistribute to lists, requires prior specic permission
and/or a fee. Request permissions from
PredictGIS’17, November 7–10, 2017, Redondo Beach, CA, USA
©2017 Association for Computing Machinery.
ACM ISBN 978-1-4503-5501-8/17/11.. .$15.00
Nowadays, most people carry a smartphone capable of determin-
ing its position via GSM, GPS or WiFi wherever they go. Several
applications already use these location measurements to passively
record the movement of people, for example to provide them with
feedback on their mobility behavior [1, 2, 20], to timely notify them
about upcoming appointments (at another location) [32], to give
location recommendations [38], or to warn them about trac inci-
dents. However, these high-level applications often only consider
the momentary location of a user. If they do mobility prediction,
they usually do not take the exact path someone might travel in
the near future into account, but instead consider the transition
probabilities from the current location to previously visited points
of interest (such as home, workplace, or the parents’ home).
In a similar manner, most vehicles are equipped with GPS sen-
sors, and recent trends in autonomous driving add cameras, radars
and laser measurement devices for environment surveillance and
position tracking. Highly accurate measurements from these sen-
sors make it possible to predict future positions of a vehicle, in
particular when combined with high-resolution maps of roads.
Between simply predicting the “next place” a person is going
to be and predicting the exact position of a vehicle, there is path
prediction. Several applications could benet from being able to
predict mobility on such an intermediate level. For example, due
to the recent trend towards increasingly multi-modal mobility be-
havior [23], it would be great to notify people about alternative
travel options along their traveled path, such as other cars driving
the same route (for carpooling), trains about to leave, or bikeshar-
ing stations with bikes available. Another prominent application
concerns all sorts of advertising (for restaurants, gas stations, etc.),
which requires a high prediction accuracy to timely suggest up-
coming places. Accurate mobility prediction is also valuable when
location tracking becomes unavailable, for example inside build-
ings, tunnels, or urban canyons.
In this paper, we summarize current research trends from the
various elds involved in dierent levels of mobility prediction,
and present ideas and suggestions where these elds could be headed
in the future. We conclude by envisioning an integrated system
that is able to predict mobility at various levels of detail, for any
future point in time.
PredictGIS’17, November 7–10, 2017, Redondo Beach, CA, USA Dominik Bucher
Arguably, volunteered geographic information (VGI) [14] is best
known for its use in open mapping applications, most notably Open-
StreetMap [13]. Next to the road network, OpenStreetMap con-
tains data about all kinds of spatial objects, e.g., points of inter-
est, land use zones, public transport routes, or bikesharing stations.
Goodchild [11] notes that not only is volunteered geographic infor-
mation relatively cheap to produce, but that it also is available to
almost everyone, which makes it an interesting base for the com-
putation and improvement of mobility predictions.
2.1 Road Network Data
Traditionally, road networks are formalized as graphs G=(V,E),
connecting vertices viVwith edges (vi,vj) E. Research
on autonomous vehicles recently initiated a shift towards high-
resolution maps [30, 35]. While street data has been collected using
GPS sensors in cars for a while now, we can expect high-resolution
maps for a large portion of the road network in the near future, be-
cause more and more cars are being equipped with high-precision
data collection tools, as these are required for elaborate forms of
driver assistance. Though most data collection currently happens
within proprietary bounds, it can be safely assumed that similar
tools will be freely available soon.
Thus, it will be essential to provide a repository for volunteered
high-resolution maps. Such a repository will require new data stor-
age formats, where road network vertices and edges are enhanced
by their exact extents and other street peculiarities. This will even-
tually allow very ne-grained movement predictions, which are
especially important for all forms of autonomous mobility.
In a similar manner, as indoor tracking improves and measuring
of indoor spaces (e.g., using light detection and ranging (LIDAR)
systems) becomes widely feasible, VGI services for storage and re-
trieval of indoor data should be developed and provided [10].
2.2 Other Geographic Data
Most prediction algorithms work on clustered user location his-
tories (GPS trackpoints, social media checkins, etc.), and do not
take into account other geographical information (except the road
network). Aggregated over all users, such location histories allow
predicting the movement of other users in similar situations [15].
While these algorithms show good prediction accuracies, they are
not able to predict the exact path someone might take, which is de-
pending on the (geographical) context a person is in. For example,
a bicycle rider is more likely to choose a less dangerous road along
a park, a street with a smaller inclination, or to take a bus in case
of rain, even though this might prolong the travel duration signi-
cantly [cf. 3]. Such geographic data is not necessarily available in a
graph-based form, but can consist of vector or raster data, e.g., de-
picting landuse data, marking dangerous spots or air quality. These
heterogeneous data sources either need to be integrated into exist-
ing models of mobility (e.g., using a spatio-temporal form of map
algebra [19]), or new models should be developed that incorporate
contextual geographical information.
2.3 Real-time Data
Fine-grained mobility prediction will also largely be driven by real-
time data. Vehicular ad-hoc networks (VANET) allow vehicles to
share information about their movement and surroundings, and to
coordinate future movement [24]. Adding real-time data about the
trac state, potential obstacles (such as construction sites or acci-
dents), and special events to VANET data will strongly inuence
the mobility behavior of people and vehicles. As such, prediction
models need to be able to quickly adapt to new circumstances, i.e.,
they need to be continuously updated.
The location prediction problem can be formulated as follows. Given
a movement trajectory
T=⟨(lon1,lat1,t1),(lon2,lat2,t2), . .., (lonn,latn,tn)⟩
consisting of time-stamped coordinates, we want to nd the future
path of a user, i.e., (lonm,l atm,tm)for some m>n(resp. tm>tn).
High-level movement prediction associates points of interest with
location tuples (lon,lat )and is interested in a future sequence of
points of interest. This requires clustering of user locations, which
is usually done either grid-based or using some neighborhood- or
density-based approach. For ne-grained path prediction we usu-
ally associate street intersections (i.e., road network vertices) with
a location tuple, and interpolate on the roads between. This means
that we are not strictly interested in a sequence of future locations,
but rather a function f(T,t)which assigns a location (lon,lat )to
every t>tn. In practice, most algorithms yield a tree of possible
future paths, where each edge is associated with a certain probabil-
ity that the person travels this route. f(T,t)would simply assign
the location of the most probable path to time t.
A very closely related eld is map matching, which uses similar
mobility models to assess which roads a person most likely took. In
the following sections, we will refer to both movement prediction
as well as map matching.
3.1 Motion Functions and Heuristic Methods
On a very low level, linear motion functions describe the move-
ment of physical objects well. While not directly applicable to pre-
dictions in a road network (due to the fork dilemma, which de-
scribes the necessary turn at a T-shaped intersection [18]), they
are commonly used for indexing moving objects in databases [29].
However, with respect to the increase in location recording fre-
quencies (e.g., in autonomously driving cars), it becomes necessary
to integrate motion functions with a higher level of mobility path
Heuristic-based mobility models (and expert systems) are usu-
ally used in routing applications, where parameters (e.g., for stop
times at trac lights, left turn penalties, or acceleration) are man-
ually set, as no information about the application user is known
and the resulting route should be optimized for duration [e.g. 26].
Such models should form a basis upon which learning models can
build, e.g., by learning user-dependent parameter values.
Volunteered Geographic Information for Mobility Prediction PredictGIS’17, November 7–10, 2017, Redondo Beach, CA, USA
3.2 Markov Models and Machine Learning
A large body of literature covers the prediction of “next places”,
i.e., important spots in either a person’s life or within a certain geo-
graphical area [e.g., 5, 9, 21, 33, 39]. This focus on individual points
of interest (POIs) has several reasons: Knowing about stops at im-
portant places provides a sucient base for many applications, as
they are primarily interested in performing actions related to POIs,
and not the paths between them. In addition, many datasets are
available in such a “clustered” format, be this due to restrictions
during the data collection (e.g., mobile phone usage that is strictly
linked to cell tower ids), or due to privacy reasons (clustering user
locations into grid cells, inter-user clusters, etc.). Finally, focusing
on POIs also reduces the computational complexity of the prob-
lem, and conceptually maps very well to state-based models (e.g.,
Markov models, where each state corresponds to a single point of
interest). A variety of machine learning methods have been applied
to the next place problem, such as Markov models, Support Vector
Machines and decision trees [28].
When looking at more ne-grained mobility prediction, we nd
applications of similar methods as above (where each road junction
is treated like a point of interest or the trackpoints are aggregated
into cells) [22, 31, 37, 39]. When predicting a future route, these net-
work mobility models usually resemble graph-based routing with
adapted edge weights [18]. The same models are often also used
for map matching [27], where ambiguous trackpoints need to be
associated with a single road segment.
A less common approach is the use of lters for short-term mo-
bility prediction. Kalman or particle lters have been applied suc-
cessfully to location prediction in vehicular ad-hoc networks [17].
In contrast to above methods, they are able to predict positions of
objects on a very ne-grained level, as they do not have to consider
cells or location clusters.
3.3 Deep Neural Networks
Many applications of neural network models to mobility predic-
tion are based on a similar “next place” problem denition [6, 7, 12,
25], reaching accuracies around 60% [7]. Liu et al. [25] incorporate
spatial and temporal context by adding time- and distance-specic
transition matrices. This is closely related to the research by We
et al. [36], which pays special attention to keep road and context
information, and to enable processing of long location sequences.
Other works in the eld of neural networks for the processing of
spatio-temporal data include movement pattern extraction for vi-
sual analysis [4] and the combination with spatio-temporal graphs
that encode high-level expert knowledge [16].
While the work by Jain et al. [16] is not directly related to human
mobility prediction within a city or geographical area, studying
spatio-temporal dynamics of humans is a hot topic in computer
vision and neural networks research [8, 34].
We expect future research on the use of neural networks for mo-
bility prediction to increase, as neural networks are well suited to
process spatial data. Convolutional networks are commonly used
to detect patterns in images, which could directly be used to pro-
cess geographical raster data, or to assess rasterized street net-
Figure 1 summarizes the dierent parts required for mobility pre-
diction as presented in this paper. As mobility predictions at level 3
(considering sequences of points of interest) have mostly been in
the focus of previous research, there is a wide variety of methods
available which have also been successfully deployed in commer-
cial applications (e.g., Google Maps notifying you when it is time to
leave work to be home on time). These high-level predictions usu-
ally do not make heavy use of volunteered geographical informa-
tion, except to train models across users, e.g., by using Foursquare
checkins to learn common movement patterns (“after having din-
ner at a restaurant, people commonly go home”).
We expect the lower two levels to be more in the attention of re-
searchers in the near future, not only because the increased amount
of data and increased tracking accuracies allow us to train more
ne-grained models, but also because new forms of mobility re-
quire path predictions (e.g., for real-time multi-modal transport
suggestions), in the case of autonomous mobility even the pre-
diction of the exact position on the road (both for the vehicle it-
self, as well as for any other moving object in its vicinity). While
the predictions at level 1 are mostly based on physical models of
movement (however, taking into account restrictions as given by
high-resolution maps), level 2 has to incorporate the biggest wealth
of volunteered geographic information: street data, landuse zones,
real-time trac updates, and so on.
We ultimately envision a system (based on future research in
the eld) that integrates the dierent levels of mobility prediction,
and thus can be used to compute a set of probable locations for any
given time in the future.
Acknowledgements. This research was supported by the Swiss
National Science Foundation (SNF) under the National Research
Program NRP71 “Managing Energy Consumption” and by the Com-
mission for Technology and Innovation (CTI) within the Swiss Com-
petence Center for Energy Research (SCCER) Mobility.
[1] Jing Bie, Marcel Bijlsma, Gregor Broll, Hu Cao, Anders Hjalmarsson, Frances
Hodgson, Paul Holleis, et al. 2012. Move better with tripzoom. International
journal on advances in life sciences 4, 3&4 (2012), 125–135.
[2] Dominik Bucher, Francesca Cellina, Francesca Mangili, Martin Raubal, Roman
Rudel, Andrea Emilio Rizzoli, and Omar Elabed. 2016. Exploiting tness apps for
sustainable mobility-challenges deploying the goeco! app. ICT for Sustainability
(ICT4S) (2016).
[3] Dominik Bucher, David Jonietz, and Martin Raubal. 2017. A Heuristic for Multi-
modal Route Planning. In Progress in Location-Based Services 2016. Springer, 211–
[4] Zhenghao Chen, Jianlong Zhou, and Xiuying Wang. 2017. Visual Analytics of
Movement Pattern Based on Time-Spatial Data: A Neural Net Approach. arXiv
preprint arXiv:1707.02554 (2017).
[5] Sung-Bae Cho. 2016. Exploiting machine learning techniques for location recog-
nition and prediction with smartphone logs. Neurocomputing 176 (2016), 98–
[6] Alexandre De Brébisson, Étienne Simon, Alex Auvolat, Pascal Vincent, and
Yoshua Bengio. 2015. Articial neural networks applied to taxi destination pre-
diction. arXiv preprint arXiv:1508.00021 (2015).
[7] Vincent Etter, Mohamed Kafsi, and Ehsan Kazemi. 2012. Been there, done that:
What your mobility traces reveal about your behavior. In Mobile Data Challenge
by Nokia Workshop, in conjunction with Int. Conf. on Pervasive Computing.
[8] Katerina Fragkiadaki, Sergey Levine, Panna Felsen, and Jitendra Malik. 2015. Re-
current network models for human dynamics. In Proceedings of the IEEE Inter-
national Conference on Computer Vision. 4346–4354.
[9] Sébastien Gambs, Marc-Olivier Killijian, and Miguel Núñez del Prado Cortez.
2012. Next place prediction using mobility markov chains. In Proceedings of the
PredictGIS’17, November 7–10, 2017, Redondo Beach, CA, USA Dominik Bucher
Figure 1: The three-level mobility prediction system. Each level favors dierent inputs and models, and passes its output
mobility prediction to the next lower level.
First Workshop on Measurement, Privacy, and Mobility. ACM, 3.
[10] Marcus Goetz and Alexander Zipf. 2011. Extending OpenStreetMap to indoor
environments: bringing volunteered geographic information to the next level.
Urban and Regional Data Management: Udms Annual 2011 (2011), 47–58.
[11] Michael F Goodchild. 2007. Citizens as sensors: the world of volunteered geog-
raphy. GeoJournal 69, 4 (2007), 211–221.
[12] Sedef Gunduz, Uraz Yavanoglu, and Seref Sagiroglu. 2013. Predicting next lo-
cation of Twitter users for surveillance. In Machine Learning and Applications
(ICMLA), 2013 12th International Conference on, Vol. 2. IEEE, 267–273.
[13] Mordechai Haklay and Patrick Weber. 2008. Openstreetmap: User-generated
street maps. IEEE Pervasive Computing 7, 4 (2008), 12–18.
[14] Christian Heipke. 2010. Crowdsourcing geospatial data. ISPRS Journal of Pho-
togrammetry and Remote Sensing 65, 6 (2010), 550–557.
[15] Qunying Huang. 2017. Mining online footprints to predict users next location.
International Journal of Geographical Information Science 31, 3 (2017), 523–541.
[16] Ashesh Jain, Amir R Zamir, Silvio Savarese, and Ashutosh Saxena. 2016.
Structural-RNN: Deep learning on spatio-temporal graphs. In Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition. 5308–5317.
[17] Raj K Jaiswal and CD Jaidhar. 2016. Location prediction algorithm for a nonlinear
vehicular movement in vanet using extended kalman lter. Wireless Networks
(2016), 1–16.
[18] Hoyoung Jeung, Man Lung Yiu, Xiaofang Zhou, and Christian S Jensen. 2010.
Path prediction and predictive range querying in road network databases. The
VLDB JournalThe International Journal on Very Large Data Bases 19, 4 (2010),
[19] David Jonietz and Dominik Bucher. 2017. Towards an Analytical Framework for
Enriching Movement Trajectories with Spatio-Temporal Context Data. Interna-
tional Conference on Geographic Information Science (AGILE 2017) (2017).
[20] Antti Jylhä, Petteri Nurmi, Miika Sirén, Samuli Hemminki, and Giulio Jacucci.
2013. Matkahupi: a persuasive mobile application for sustainable mobility. In
Proceedings of the 2013 ACM conference on Pervasive and ubiquitous computing
adjunct publication. ACM, 227–230.
[21] Faina Khoroshevsky and Boaz Lerner. 2016. Human Mobility-Pattern Discov-
ery and Next-Place Prediction from GPS Data. In IAPR Workshop on Multimodal
Pattern Recognition of Social Signals in Human-Computer Interaction. Springer,
[22] John Krumm. 2016. A markov model for driver turn prediction. (2016).
[23] Tobias Kuhnimhof, Ralph Buehler, and Joyce Dargay. 2011. A new generation:
travel trends for young Germans and Britons. Transportation Research Record:
Journal of the Transportation Research Board 2230 (2011), 58–67.
[24] Fan Li and Yu Wang.2007. Routing in vehicular ad hoc networks: A survey. IEEE
Vehicular technology magazine 2, 2 (2007).
[25] Qiang Liu, Shu Wu, Liang Wang, and Tieniu Tan. 2016. Predicting the Next
Location: A Recurrent Model with Spatial and Temporal Contexts.. In AAAI. 194–
[26] Dennis Luxen and Christian Vetter. 2011. Real-time routing with OpenStreetMap
data. In Proceedings of the 19th ACM SIGSPATIAL International Conference on
Advances in Geographic Information Systems (GIS ’11). ACM, New York, NY, USA,
[27] Paul Newson and John Krumm. 2009. Hidden Markov map matching through
noise and sparseness. In Proceedings of the 17th ACM SIGSPATIAL international
conference on advances in geographic information systems. ACM, 336–343.
[28] Binh T Nguyen, Nhan V Nguyen, Nam T Nguyen, and My Huynh T Tran. 2017. A
potential approach for mobility prediction using GPS data. In Information Science
and Technology (ICIST), 2017 Seventh International Conference on. IEEE, 45–50.
[29] Simonas Saltenis and Christian S Jensen. 2002. Indexing of moving objects for
location-based services. In Data Engineering, 2002. Proceedings. 18th International
Conference on. IEEE, 463–472.
[30] Andreas Schindler. 2013. Vehicle self-localization with high-precision digital
maps. In Intelligent Vehicles Symposium (IV), 2013 IEEE. IEEE, 141–146.
[31] Reid Simmons, Brett Browning, Yilu Zhang, and Varsha Sadekar. 2006. Learn-
ing to predict driver route and destination intent. In Intelligent Transportation
Systems Conference, 2006. ITSC’06. IEEE. IEEE, 127–132.
[32] Timothy Sohn, Kevin Li, Gunny Lee, Ian Smith, James Scott, and William Gris-
wold. 2005. Place-its: A study of location-based reminders on mobile phones.
UbiComp 2005: Ubiquitous Computing (2005), 903–903.
[33] William Su, S-J Lee, and Mario Gerla. 2000. Mobility prediction in wireless net-
works. In MILCOM 2000. 21st Century Military Communications Conference Pro-
ceedings, Vol. 1. IEEE, 491–495.
[34] Graham W Taylor, Leonid Sigal, David J Fleet, and Georey E Hinton. 2010. Dy-
namical binary latent variable models for 3d human pose tracking. In Computer
Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on. IEEE, 631–638.
[35] Christopher KH Wilson, Seth Rogers, and Shawn Weisenburger. 1998. The po-
tential of precision maps in intelligent vehicles. In IEEE International Conference
on Intelligent Vehicles. 419–422.
[36] Fan Wu, Kun Fu, Yang Wang, Zhibin Xiao, and Xingyu Fu. 2017. A Spatial-
Temporal-Semantic Neural Network Algorithm for Location Prediction on Mov-
ing Objects. Algorithms 10, 2 (2017), 37.
[37] Guangtao Xue, Zhongwei Li, Hongzi Zhu, and Yunhuai Liu. 2009. Trac-known
urban vehicular route prediction based on partial mobility patterns. In Parallel
and Distributed Systems (ICPADS), 2009 15th International Conference on. IEEE,
[38] Mao Ye, Peifeng Yin, and Wang-Chien Lee. 2010. Location Recommendation for
Location-based Social Networks. In Proceedings of the 18th SIGSPATIAL Interna-
tional Conference on Advances in Geographic Information Systems (GIS ’10). ACM,
New York, NY, USA, 458–461.
[39] Weimin Zheng, Xiaoting Huang, and Yuan Li. 2017. Understanding the tourist
mobility using GPS: Where is the next place? Tourism Management 59 (2017),
... In [2] three levels of mobility prediction are considered: (1) object position; (2) path prediction; and (3) next place prediction. In levels 1 and 2, usually, the models learned from raw trajectories obtained by Global Positioning System (GPS) devices, and their predictions consider the movement of the object. ...
... We can also model our problem as a sequence of sensors' labels, allowing us to apply RNN to EST prediction. Thus, we aim to investigate the following questions: (1) Can RNN adapt to incompleteness and sparsity of EST? (2) How different data imputation strategies impact RNN models? (3) How to access the quality of predictions given the fact that the ground truth could be missed and thus unknown? ...
... We executed each experiment 3 times and computed the average. For the models that use a clustering-based strategy, we vary the value of eps in the range (1,2,3,4,5) km. ...
Full-text available
This paper proposes a deep multi-task learning framework to predict the next location from trajectories that are captured by external sensors (e.g., traffic surveillance cameras, or speed radars). The reported positions in such trajectories are sparse, due to the sparsity of the sensor distribution, and incomplete, because the sensors may fail to register the passage of objects. In this framework, we propose different preprocessing steps to align the trajectories representation and cope with a missing data problem. The multi-task learning approach is based on Recurrent Neural Networks. It utilizes both time and space information in the training phase to learn more meaningful representations, which boosts the learning performance of location prediction. The multi-task learning model, together with the preprocessing step, substantially improves the prediction performance. We conduct several experiments using a real dataset, and they demonstrate the validity of our multi-task learning model in terms of accuracy of 85.20%, which is more than 20% better than using a single-task learning model.
... In fact, predicting certain aspects of our movement behaviour is important to numerous application and serves various purposes. One can predict movement on different levels [9]. On a very high level, mobility prediction is for example important to determine bottlenecks in transportation systems and plan appropriate transportation infrastructure [10], to provide operational guidance in disaster situations [11], or to detect potentially dangerous upcoming events [12]. ...
... A hierarchical approach could also use different prediction models and methods depending on the spatial resolution and the task at hand (cf. [9]), and incorporate an automated approach for determining the optimal number of levels and their spatial scales. Furthermore, our work could be extended by testing additional classification algorithms and methods. ...
Full-text available
For next place prediction, machine learning methods which incorporate contextual data are frequently used. However, previous studies often do not allow deriving generalizable methodological recommendations, since they use different datasets, methods for discretizing space, scales of prediction, prediction algorithms, and context data, and therefore lack comparability. Additionally, the cold start problem for new users is an issue. In this study, we predict next places based on one trajectory dataset but with systematically varying prediction algorithms, methods for space discretization, scales of prediction (based on a novel hierarchical approach), and incorporated context data. This allows to evaluate the relative influence of these factors on the overall prediction accuracy. Moreover, in order to tackle the cold start problem prevalent in recommender and prediction systems, we test the effect of training the predictor on all users instead of each individual one. We find that the prediction accuracy shows a varying dependency on the method of space discretization and the incorporated contextual factors at different spatial scales. Moreover, our user-independent approach reaches a prediction accuracy of around 75%, and is therefore an alternative to existing user-specific models. This research provides valuable insights into the individual and combinatory effects of model parameters and algorithms on the next place prediction accuracy. The results presented in this paper can be used to determine the influence of various contextual factors and to help researchers building more accurate prediction models. It is also a starting point for future work creating a comprehensive framework to guide the building of prediction models.
Full-text available
Time-Spatial data plays a crucial role for different fields such as traffic management. These data can be collected via devices such as surveillance sensors or tracking systems. However, how to efficiently an- alyze and visualize these data to capture essential embedded pattern information is becoming a big challenge today. Classic visualization ap- proaches focus on revealing 2D and 3D spatial information and modeling statistical test. Those methods would easily fail when data become mas- sive. Recent attempts concern on how to simply cluster data and perform prediction with time-oriented information. However, those approaches could still be further enhanced as they also have limitations for han- dling massive clusters and labels. In this paper, we propose a visualiza- tion methodology for mobility data using artificial neural net techniques. This method aggregates three main parts that are Back-end Data Model, Neural Net Algorithm including clustering method Self-Organizing Map (SOM) and prediction approach Recurrent Neural Net (RNN) for ex- tracting the features and lastly a solid front-end that displays the results to users with an interactive system. SOM is able to cluster the visiting patterns and detect the abnormal pattern. RNN can perform the predic- tion for time series analysis using its dynamic architecture. Furthermore, an interactive system will enable user to interpret the result with graph- ics, animation and 3D model for a close-loop feedback. This method can be particularly applied in two tasks that Commercial-based Promotion and abnormal traffic patterns detection.
Conference Paper
Full-text available
We provide a novel algorithm for the discovery of mobility patterns and prediction of users’ destination locations, both in terms of geographic coordinates and semantic meaning. We did not use any semantic data voluntarily provided by a user, and there was no sharing of data among the users. An advantage of our algorithm is that it allows a trade-off between prediction accuracy and information. Experimental validation was conducted on a GPS dataset collected in the Microsoft Research Asia GeoLife project by 168 users in a period of over five years.
Conference Paper
Full-text available
Although there are numerous studies in which movement trajectories were analysed based on their geometry alone, a more holistic interpretation requires their contextual setting to be incorporated into the analytical process as well. Among other influences, for instance, several attributes of the underlying physical space can have an effect on movement. Thus, a critical task is to semantically enrich movement trajectories with attribute values derived from one or more underlying layers with spatio-temporal data. This process, however, is not always trivial due to the fact that currently, there exists no comprehensive framework or toolset for such purposes. Further, communicating the exact procedure which was used in a study is impeded by a lack of formal terminology required to unambiguously define the involved methods and operators. In this paper, we address these issues and conceptualize an analytical framework for the process of enriching movement trajectories with spatio-temporal context data from underlying data layers, as well as propose a terminology to explicitly define the spatio-temporal scope of the analysis. We demonstrate the validity of our concept on a practical example.
Full-text available
Location prediction has attracted much attention due to its important role in many location-based services, such as food delivery, taxi-service, real-time bus system, and advertisement posting. Traditional prediction methods often cluster track points into regions and mine movement patterns within the regions. Such methods lose information of points along the road and cannot meet the demand of specific services. Moreover, traditional methods utilizing classic models may not perform well with long location sequences. In this paper, a spatial-temporal-semantic neural network algorithm (STS-LSTM) has been proposed, which includes two steps. First, the spatial-temporal-semantic feature extraction algorithm (STS) is used to convert the trajectory to location sequences with fixed and discrete points in the road networks. The method can take advantage of points along the road and can transform trajectory into model-friendly sequences. Then, a long short-term memory (LSTM)-based model is constructed to make further predictions, which can better deal with long location sequences. Experimental results on two real-world datasets show that STS-LSTM has stable and higher prediction accuracy over traditional feature extraction and model building methods, and the application scenarios of the algorithm are illustrated.
Full-text available
Current popular multi-modal routing systems often do not move beyond combining regularly scheduled public transportation with walking, cycling or car driving. Seldom included are other travel options such as carpooling, carsharing, or bikesharing, as well as the possibility to compute personalized results tailored to the specific needs and preferences of the individual user. Partially, this is due to the fact that the inclusion of various modes of transportation and user requirements quickly leads to complex, semantically enriched graph structures, which to a certain degree impede downstream procedures such as dynamic graph updates or route queries. In this paper, we aim to reduce the computational effort and specification complexity of personalized multi-modal routing by use of a preceding heuristic, which, based on information stored in a user profile, derives a set of feasible candidate travel options, which can then be evaluated by a traditional routing algorithm. We demonstrate the applicability of the proposed system with two practical examples.
Conference Paper
Full-text available
The large interest in analyzing one's own fitness led to the development of more and more powerful smartphone applications. Most are capable of tracking a user's position and mode of locomotion, data that do not only reflect personal health, but also mobility choices. A large field of research is concerned with mobility analysis and planning for a variety of reasons, including sustainable transport. Collecting data on mobility behavior using fitness tracker apps is a tempting choice, because they include many of the desired functions, most people own a smartphone and installing a fitness tracker is quick and convenient. However, as their original focus is on measuring fitness behavior, there are a number of difficulties in their usage for mobility tracking. In this paper we denote the various challenges we faced when deploying GoEco! Tracker (an app using the Moves R fitness tracker to collect mobility measurements), and provide an analysis on how to best overcome them. Finally, we summarize findings after one month of large scale testing with a few hundred users within the GoEco! living lab performed in Switzerland.
Spatial and temporal contextual information plays a key role for analyzing user behaviors, and is helpful for predicting where he or she will go next. With the growing ability of collecting information, more and more temporal and spatial contextual information is collected in systems, and the location prediction problem becomes crucial and feasible. Some works have been proposed to address this problem, but they all have their limitations. Factorizing Personalized Markov Chain (FPMC) is constructed based on a strong independence assumption among different factors, which limits its performance. Tensor Factorization (TF) faces the cold start problem in predicting future actions. Recurrent Neural Networks (RNN) model shows promising performance comparing with PFMC and TF, but all these methods have problem in modeling continuous time interval and geographical distance. In this paper, we extend RNN and propose a novel method called Spatial Temporal Recurrent Neural Networks (ST-RNN). ST-RNN can model local temporal and spatial contexts in each layer with time-specific transition matrices for different time intervals and distance-specific transition matrices for different geographical distances. Experimental results show that the proposed ST-RNN model yields significant improvements over the competitive compared methods on two typical datasets, i.e., Global Terrorism Database (GTD) and Gowalla dataset.
Conference Paper
In this paper, we investigate the problem of user movement prediction from historical location data. We create an Android application, namely Movement Predictor, that can help to collect location data from registered users by Global Positioning System (GPS) signals. We analyze different kinds of feature vectors and compare three supervised learning models: Markov model, Support Vector Machine (SVM), and decision tree. The experiments show that SVM model can achieve the highest performance (with an accuracy 92%) in comparison with other approaches.