See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308810260

Impacts of Traffic Conditions on the Performance of Road Freight Transport

Conference Paper · September 2015

DOI: 10.1109/ITSC.2015.472

CITATIONS READS
9 497

3 authors, including:

Gideon Mbiydzenyuy Johan Holmgren

University of Borås Malmö University
34 PUBLICATIONS 97 CITATIONS 73 PUBLICATIONS 780 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Gideon Mbiydzenyuy on 25 July 2017.

The user has requested enhancement of the downloaded file.

 Impacts of traffic conditions on the performance of road freight
transport
Ksenia Sigakova1 , Gideon Mbiydzenyuy2 , and Johan Holmgren3

Abstract— The efficiency of road transport is typically influ- the execution of transport could help improve planning road
enced by factors such as, weather, choice of road, and time freight transport.
of day, and day of the week. Knowledge about interactions Data processing involves retrieving useful information
between different traffic- and transport related factors and their
influence on the execution of transport is important in transport through data crunching techniques including aggregation,
planning. The purpose of this paper is to study the impact of classification, filtration, querying, and synthesis. Such tech-
different factors on the performance of road transport. We niques demand advanced and complex computational meth-
aim to contribute to improved transport planning by analysing ods and tools. For example, in their work related to real-
traffic and transport data obtained from different sources in time weather data, Mathew et. al [1] showed the degree of
order to support data driven decision making.
Through review of existing literature and discussions with
complexity in the processing of different types of data from
a Swedish road transport operator, we identified factors that various sources to synthesize useful information about the
could be relevant to consider when planning a transport, e.g., weather for individual local neighbourhoods. In particular,
related to weather, location of roads where the transport will the ability to determine the extent to which large amount
take place, and planned time of the transport. As a result of of different types of weather data influences the execution
variation in size, type and volume of the data representing these
factors, suitable machine learning algorithms were selected,
of transport in given road links, poses a challenge. In
such as Decision Stump, M5 model tree, M5 regression tree, many cases, data needs to be aggregated, for example, for
RepTree, M5 rules, and linear regression in order to study 10 minutes periods [2] to reflect the constant changing
the data. Our experimental results illustrate the complexity weather conditions. For location data, such as GPS data,
associated to the performance of road transport systems mainly huge numbers of records are produced, particularly if data
because of the dependency between the choices of influencing
factors and geographic location of the road segment.
is collected at short sampling periods, say every 5th second
[3], [4]. Relevant techniques for synthesizing GPS data under
I. I NTRODUCTION various conditions can reveal recurring events, e.g., related
to incidents, speed, and travel times.
The efficiency of road transport is typically influenced by The purpose of this paper is to study the impact of
factors, such as, weather, choice of road, and time of day, different factors on the performance of road transport. We
and day of the week. One way to understand the conditions aim to contribute to improve transport planning by analysing
causing these factors is by analysing the underlying data traffic and transport data obtained from different sources in
about such factors. Commercial transport organisations and order to support data driven decision making. Data driven
government agencies responsible for road infrastructure al- decision making, e.g., planning of transport delivery time
ready collects, store and process different types of data with using historical data, can be improved by combining data
the help of different Information and Communication Tech- describing different factors such as weather and location,
nologies (ICT). For instance, temperature measurements at traffic volumes, and travel patterns during a day or week.
given locations, wind speed, and precipitation, are examples We reviewed existing literature and discussed possible
of weather data collected using probes. Such data as weather factors with a Swedish road transport operator. This enabled
data can be characterised as random continuously changing us to identify relevant factors to consider when planning a
data, rich in variety and volume, making it challenging to transport, e.g., related to weather, location of roads where the
process the data and generate useful information, such as transport will take place, and planned time of the transport.
travel times.Traffic and transport data processing is neces- Conditions representing these factors are captured through
sary to understand how different factors affects transport different data sources; weather conditions through temper-
operations. Knowledge about interactions between different ature, precipitation, wind direction, location through GPS
traffic- and transport related factors and their influence on data, travel patterns through records of traffic during different
times of the day, and so on. The data collected was then pre-
1 kseniasigakova@gmail.com, Former M.Sc. Student, Department of
processed (e.g., filtering outliers) in order to conform with
Computer Science and Engineering, Blekinge Institute of Technology, selected choices of machine learning techniques.
Karlshamn, Sweden
2 gideon.mbiydzenyuy@netport.se, Karlshamn, Sweden; NetPort Science The ability to consider the impact of different types of
Park AB, Karlshamn; Former researcher at Department of Computer Science traffic and transport information is crucial for the successful
and Engineering, Blekinge Institute of Technology execution of transport [5], [6]. In addition, previous research
3 johan.holmgren@mah.se, Department of Computer Science, Malmö
University, Malmö SE-205 06, Sweden; Department of Computer Science show that better travel time estimates can be obtained when
and Engineering, Blekinge Institute of Technology, Karlshamn, Sweden several types of traffic and transport information are jointly
considered. For example, [6], [7], [8] consider day-of-week of traffic-related factors on the performance of road freight
patterns, [9], [10], [11] combine traffic data with weather transport. In particular, we used the following supervised
data. Different types of computational models have been learning algorithms:
employed in order to analyse different factors influencing • Decision Stump (a tree with the only node indicating
road transport. By means of an integrated nonparametric the most significant factor),
model, KNN-integrated, Qiao et al. [12] showed that there • M5 pruned model tree (a model tree built using M5
is a potential in using computational based models in order algorithm whose tree leaves are based on multivariate
to study the effects of weather in short term travel time linear models),
prediction. A statistical model for travel time estimation • M5 pruned regression tree (a regression tree built using
that considers the day of the week, season, year, holidays, M5 algorithm),
and vacation periods is suggested [8]. Another travel time • RepTree (a regression tree built using variance reduction
prediction model makes use of hour of the day, the day of and reduced-error pruning),
the month and the day of the week as input parameters • M5 Rules (rules obtained from a model tree),
[6]. Location based data collected with the help of GPS • Linear Regression (the model in the form of an equa-
have been used in order to calculate travel times using tion), and
two different models: the speed model and the proportional • ZeroR (predicts the mean of the output values- required
model [13] with the proportional model indicating a better for comparison)
performance. Beside the studies mentioned above, we could
not find extensive research work with focus on the analysis of These algorithms were selected due to their ability to build
multiple factors in order to estimate transport performance, prediction models that can be used to estimate continuous
e.g, based on travel times. values (in our case average speed for traversing a road and
In the current study, we use machine learning algorithms in hence travel time) from the values of a set of attributes.
order to study the effect of multiple factors on road transport. Moreover, the prediction models generated with these algo-
As a result of variation in size, type and volume of the rithms are typically straightforward to analyse.
data representing these factors, suitable machine learning Initially, we reviewed the literature and performed open-
algorithms were selected, such as Decision Stump, M5 model ended interviews with management staff at a Swedish freight
tree, M5 regression tree, RepTree, M5 rules, and linear transport operator in order to identify which factors typi-
regression in order to study the data. Machine learning tech- cally have an influence on the performance of road freight
niques are employed because they facilitate the possibility to transport; in particular on the travel time or traveling speed,
detect and describe patterns in large data sets. In addition, which we consider as two of the most important measures
previous research indicates that machine learning techniques of transport performance. Based on the studied literature, the
provide useful insights into the study of road transport interview, and the type of data available, we selected a set of
performance, see, e.g., [14], [6], [5]. To demonstrate this in factors that we analysed in detail in an experiment including
an experiment, we concentrated on weather and time patterns two geographically separated road segments with different
for specific road segments. We designed different experiment characteristics. For the selected factors, we collected various
scenarios specific to transport using heavy goods vehicles. types of data such as weather data, road network data, and
The relative impacts of different factors on travel times were historical GPS trajectories for heavy goods vehicles. This
compared in an effort to assess how such factors may affect data was then filtered and processed in order to eliminate
travel times. errors, and to convert data into a form that was suitable for
The results from our experiment scenarios indicates which further analysis (the data management is discussed in more
factors accounted for significant impact on the travel time detail in Section IV).
of heavy goods vehicles. Further, result obtained from ex- For each of the road segments, we generated prediction
periments illustrate the dependency between the choices models by employing the algorithms listed above imple-
of influencing factors and geographic location of the road mented in WEKA (Waikato Environment for Knowledge
segment. For a practitioner, e.g., road traffic planner, this Analysis) [15]. The WEKA software was chosen because
study accentuates the need for dynamic calibration of models its implementation of the algorithms provided additional
built to incorporate multiple traffic factors. Such dynamic possibilities, such as pruning and smoothing. This makes
calibration may vary according to road segment type and it possible to build more accurate models, that can handle
time in question. The remainder of the paper is organised as both numeric and nominal attributes, as well as missing
follows: Section II discusses the research method, III Factors values. Also, WEKA has a range of algorithms for result
influencing road freight transport, Section IV presents the evaluation. We evaluated our prediction models using 10-fold
experiments, in Sections V we discuss the result and in cross validation and the Root-Mean Squared Error (RMSE)
Section VI conclusions and future work are presented. measure, as it is a commonly used technique. Further, we
analysed the resulting prediction models by studying the
II. R ESEARCH M ETHOD suggested parameters from the algorithm. As different types
As discussed in the introduction, we used machine- of prediction algorithms select the most significant factors
learning techniques in order to study the possible impact for a given input dataset (when generating patterns), we
argue that it is possible to analyse the resulting models in multiple segment travel time along a given road, the travel
order to understand factors with a significant impact on the time for the entire transport can be estimated [4].
performance of road transport. Location-Related Factors: If a transport planner is aware
that the demand for capacity in a road segment at a particular
III. FACTORS INFLUENCING ROAD FREIGHT TRANSPORT time in a given location will be high, an easy solution will be
to use an alternative route or choose another time for delivery.
We studied scientific articles related to how different For example, this could be the case if the initial road traverses
activities may influence transport operations, e.g., transport an urban area. An urban area only permit lower speed limits,
performance [16]. Other factors are identified through open- has several traffic lights and pedestrian crossings and a
ended interviews with management staff at a Swedish trans- relatively high demand on road capacity. This could then
port organisation. Examples includes weather changes under lead to frequent vehicle speed-down and stops, congestion,
which a transport was performed, the type of road that longer traffic queues, and higher numbers of road incidents.
was chosen, the duration or travel time that was used, the If part or the entire planned road is through mountainous
loading and unloading duration, the capacity utilisation of areas, different types of challenges may come into play.
the vehicle, and the road traffic volumes. We choose to focus In mountainous areas the impact of weather conditions on
on weather-related, time-related and location-related factors traffic may be significant than in areas relatively closer to the
(considering data from different road segments). sea level. Different geographic locations differentiate routes,
Weather-Related Factors: Whether conditions can be ex- e.g., European highway, national road, and local road, traffic
pressed in terms of road surface temperature, iciness, and volumes, number of lanes, speed limits ,etc, all of which are
surrounding air temperature, etc., [9], [10], [17], [18], [19], somewhat interdependent. Therefore the conditions of the
[20]. Precipitation, wind direction, speed and in some cases road in relation to its location will lead to different effects
images from cameras make it possible to understand weather in the performance of a road transport operation.
conditions [20]. Weather information has been shown to
impact travel time in many ways, including traffic demand, IV. E XPERIMENT
traffic safety and, capacity [2]. Moreover, weather has an ef- In the experiment design, data was collected for traffic
fect on traffic flow relationships, such as the flow-occupancy conditions related to weather, location and time. The tem-
and speed-flow relationships [2]. A recent study investigates poral resolution of the weather data was within 0.5hours to
the impact of the rate of precipitation and temperature levels 2 hours interval, while the spatial resolution covered two
on travel times in the London metropolitan area [10]. The road segments (Segment 1: 21.2 km, Segment 2: 66.4 km)
results indicate that the effect of precipitation on travel time within a 35 km radius, approximately. The data was then
may be significant, depending on the type and intensity of pre-processed, e.g., by filtering outliers, in order to conform
precipitation. In addition, a small impact on travel time was with selected choices of machine learning techniques (where
associated with changes in temperature. In another study, the this was necessary). The process of data filtration reduces
impact of weather (wind speed and rainfall) among other noisy data but also simplifies data complexity. This allows
things is investigated with the help of regression tree models us to focus on data concerning the most relevant factors to
[9]. The focus of that study is to predict the vehicle speed by our study, e.g., removing GPS location with faulty speed
incorporating different regression factors. Other studies focus values and isolating GPS data indicating low vehicle speeds
on accounting for the impact of different precipitation types on road links where vehicles are expected to travel at higher
on free flow speed of the vehicles [17], [18]. A proposed speeds. An overview of the identified set of traffic conditions
loglinear travel time prediction model considers day-of-week is described in detail in [21]. The chosen traffic conditions
patterns by using different coefficients for different days [7]. also made it possible to collect data that can enable their
Time-Related Factors: There are many activities involved influence on travel time to be determined. For instance, GPS
in road transport in which time plays an important role. data collected along a road connect location coordinates with
This is because in many cases the cost functions for many time stamps, from which time related factors such as season,
activities is computed with a time component. For example, the day of the week, and the part of the day can be derived,
the longer it takes to deliver products to a customer, or and then the weather conditions for that particular time and
to load/unload a HGV, the more likely it is that the cost region can be inferred.
will get higher. In addition, time also plays an important We used historical data describing the movement of two
role in the utilisation of the road infrastructure because the heavy goods vehicles (HGVs), which were operating in the
demand for road capacity turns to show high variability in southern region of Sweden. The GPS data, which was pro-
some periods of the day (mornings and evenings), days of vided by a Swedish road freight transport operator, describes
the week (Sundays and Fridays) or even seasons (summer vehicle movement during a period of 2 years. In addition,
and winter). As a result of this variation, the performance we used weather data (from the Swedish national road
of transport is affected in different ways, e.g., variations in weather monitoring system, VViS) and a detailed description
congestion and delivery time [9], [7]. Time stamps from of the Swedish road network (from the Swedish national
vehicle signatures collected along a road segment make it network database, NVDB), each of which was provided by
possible to estimate the segment travel time. By combining the Swedish Transport Administration (Trafikverket).
 We imported the GPS data and the road transport network segments. The results for Segment 1 are summarised in Table
into Google earth software [22], which made it possible II and denoted by ”o”, indicating the models (rows) and
to visually observe and study the GPS data. Based on our the factors investigated(columns). It can be seen that the air
visualisation of the GPS data, we identified, and selected temperature is used by all models. Furthermore, in Decision
two road segments, with the length of 21 km and 66 km, Stump, M5 regression tree and M5 Rules, the air temperature
respectively, for analysis. The segments were chosen because is the only considered factor; however, note that the idea of
they were frequently travelled by the two HGVs. Since these Decision Stump is that it should select only one attribute.
roads were different in characteristics this allowed us to According to these models, the travel time is longer in cases
implicitly consider different types of roads, and hence ge- where the air temperature is higher than 14◦ C. One reason
ographic segments, in our study. The first segment (Segment behind that can be that this is more likely summer time
1) is a part of the European highway E22, from Sölvesborg to with higher amount of vacation traffic and more camping
Karlshamn, and it is situated along the coast, and the second trailers that may result in more congestions, accidents and,
segment (Segment 2) is a part of the national inland road consequently, speed reductions. However, a closer look at the
24 from Melbystrand to Hässleholm. For each of the two input dataset revealed that there were several instances with
segments, we separated historical journeys that were made the air temperature above 14◦ C during spring and autumn.
by the two vehicles. For each of the journeys, we extracted This explains why the season was not the most influencing
the chosen (see Table I) weather and time related conditions. factor, as a calendar season does not necessarily determine
We used GPS data in order to estimate the travel time for the weather conditions. Some days in the spring or autumn
each of the observed journeys on the considered road seg- and spring could have comparatively higher temperatures
ments. This was performed by approximating the Euclidean than some days in the summer.
distances between different coordinates and the correspond- By analysing the remaining three models, we found that
ing time stamps from the GPS data resulting to an average according to RepTree the most influencing factors, situated
travel speed. on the first levels of the tree, are air temperature, precipitation
In order to analyse the collected data, we employed amount and wind direction. Linear Regression and M5 model
WEKA [15], which is a widely used data mining analysis tree confirm the influence of wind direction, but also consider
software package containing a collection of learning algo- the part of the day among the most significant factors.
rithms and tools. We used a set of algorithms (see Section The results for Segment 2 are presented in Table II and
II) to build prediction models for transforming the effect denoted by ”+”. For this road segment, the day of the week
of the studied factors on the travel time. As discussed in is a significant factor according to all models; four of the
Section III, we focused our study on weather-related, time- models use only this factor in their estimations. Even though
related and location-related traffic factors. The list of chosen M5 model tree consider other factors, their coefficients
attributes along with the corresponding range of values is are so small that they do not have much effect on the
shown in Table I. It should be mentioned, that for the part predicted values. However, according to Linear Regression,
of the day factor we chose to divide the day into four periods the wind direction and precipitation amount also seem to
(morning, afternoon, evening, and night), as we believe this have significant influences on the traveling speed.
is sufficient for capturing the main changes in daily traffic As the amount of input information was limited, we
volume patterns. used 10-fold cross validation with the root mean squared
The collected traffic data was processed in order to create error in order to evaluate the built prediction models. The
two input datasets corresponding to the two considered performance in terms of the considered algorithms for the
road segments that we further used in order to build and two road segments is illustrated in Table II under the columns
analyse prediction models. The independent variables are Experiment 1 and Experiment 2. It is important to mention
the factors in Table I. Each instance in the input data set here that even though the prediction algorithms generated
consists of one value for each of the factors in Table I. The models that partially contain different attributes, they do not
dependent variable is the travel time for the road segment differ much in the quality of estimation. One possible reason
under consideration. We used prediction models to model may have to do with correlations between attributes and the
the relationship between the independent variables and the fact that some of the attributes might have limited influence
dependent variables. As datasets describing the independent depending on the algorithmic technique.
variables could contain errors, unknown values for some By comparing the results obtained from the experiments,
attributes, and outliers, we also analysed relations between we found that for Segment 1 the most relevant factors,
the selected factors and the travel time using two-dimensional namely air temperature, wind direction, and amount of
plots in order to reduce the possibility of misinterpretation. precipitation, are related to weather, while in the second
experiment the day of the week have the biggest influence on
V. R ESULTS , travel time. This indicates that the influence of a particular
As described in Section IV, we used machine learning type of factors on the vehicle speed depends on the road
algorithms in order to analyse the relationship between location. For example, in our case the influence of weather
dependent and independent variables describing transport conditions on the road segment situated near the cost was
using different heavy goods vehicles along different road much higher than on the inland road segment.
 TABLE I
ATTRIBUTES USED IN THE EXPERIMENT, THEIR TYPES AND POSSIBLE VALUES . F OR THE NUMERICAL ATTRIBUTES , THE POSSIBLE VALUES COLUMN
SPECIFIES THE RANGE OF VALUES OBSERVED IN OUR DATA SET.

Attribute Type Possible values

surface temperature numeric (◦ C) [−19.6; 46.2]
dew point numeric (◦ C) [−27.2; 16.9]
air temperature numeric (◦ C) [−24.6; 27.8]
precipitation type nominal No precipitation, rain, freezing rain, snow, sleet, unknown type
precipitation amount numeric (mm) [0; 44.4]
maximum wind speed numeric (m/s) [0; 21.7]
average wind speed numeric (m/s) [0; 24.5]
wind direction nominal N, NE, NW, E, S, SE, SW, W
air humidity numeric (%) [1.1; 100]
season nominal winter, spring, summer, autumn
the day of the week nominal Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday
the part of the day nominal morning, afternoon, evening, night

TABLE II
FACTORS USED IN THE BUILT MODELS FOR S EGMENT 1 ( DENOTED BY O ) AND S EGMENT 2 ( DENOTED BY +).

Experiment 2 (+) RMSE

Experiment 1 (o) RMSE
maximum wind speed
precipitation amount

average wind speed

surface temperature

precipitation type
air temperature

wind direction
Model/Factors

day-of-week

air humidity
part-of-day

dew point
Season

Decision Stump + o 3.5077 2.5113

M5 model tree +o o o + +o + + o 3.5587 2.5939
M5 regression tree + o 3.5897 2.8632
RepTree +o o o o o 3.3979 2.5892
M5 Rules + o 3.5403 2.5552
Linear Regression + o o o +o + + +o 3.4532 2.8549
ZeroR 3.5811 2.8664

VI. C ONCLUDING REMARKS AND FUTURE WORK other types of road segments and to investigate dependencies
between important factors for different roads that influence
In this article we have investigated the influence of differ- travel time. Moreover, it would be interesting to conduct a
ent types of traffic factors on the travel time of HGVs. By correlation analysis between different attributes, in order to
using different machine learning techniques in the proposed further understand the impacts of groups of factors.
travel time estimation model, we built predictors, which we While in this study we focus on Weather, Time and
analysed in order to figure out which factors seem to have Location related factors, other factors of interest that could be
high influence on the travel time. The results showed that included depending on data availability and the scenario of
weather factors had the biggest impact for one of considered interest are: vehicle maintenance, road maintenance, customs
road segment (Segment 1), while the day of the week was the and border control, stops such as when refueling, and rail
most influencing factor for the other considered road segment crossings. Many of these activities influence each other and
(Segment 2). Therefore, we conclude that the factors with are implicitly connected to time, for instance, multiple stops
most significant influence on the performance of HGVs differ along a road can affect the travel time for that road. As a
for different types of road segments and different geographic result, the speed of the vehicle is affected which eventually
locations. The approach proposed in this study suggest and affects the transport performance.
demonstrate the possibility of jointly analysing the impact of
multiple factors on road freight transport. R EFERENCES
Future work includes analysing larger and more complete [1] M. Williams, D. Cornford, L. Bastin, R. Jones, and S. Parker, “Auto-
set of historical data, which is important in order to further matic processing, quality assurance and serving of real-time weather
evaluate the obtained results and to gain more insight in data,” Computers & Geosciences, vol. 37, no. 3, pp. 353–362, 2011.
[2] H. Tu, H. W. van Lint, and H. J. van Zuylen, “Impact of adverse
how different types of factors influence the travel time. It weather on travel time variability of freeway corridors,” in Transporta-
would also be interesting to conduct similar experiments on tion Research Board 86th Annual Meeting, no. 07-1642, 2007.
 [3] C. A. Quiroga and D. Bullock, “Travel time studies with global posi- [14] J. Rice and E. Van Zwet, “A simple and effective method for predicting
tioning and geographic in-formation systems: an integrated methodol- travel times on freeways,” Intelligent Transportation Systems, IEEE
ogy,” Transportation Research Part C: Emerging Technologies, vol. 6, Transactions on, vol. 5, no. 3, pp. 200–207, 2004.
no. 1, pp. 101–127, 1998. [15] I. H. Witten and E. Frank, Data Mining: Practical machine learning
[4] CIMA Partners presentation, “Evaluation of travel time data collection tools and techniques. Morgan Kaufmann, 2005.
technologies: An innovative approach for a large-scale network,” ITS [16] RHA-FreightBestPractice, “Performance management for efficient
Canada Annual Meeting, Tech. Rep., 2013. road freight operations,” Department for Transport, UK FreightBest-
[5] C.-H. Wu, J.-M. Ho, and D.-T. Lee, “Travel-time prediction with Practice Guide, Tech. Rep., 2010.
support vector regression,” Intelligent Transportation Systems, IEEE [17] G. Allain, F. Gamboa, P. Goudal, J.-N. Kien, and J.-M. Loubes, “Mod-
Transactions on, vol. 5, no. 4, pp. 276–281, 2004. eling weather conditions consequences on road trafficking behaviors,”
[6] L. P. Masiero, M. A. Casanova, and M. T. M. de Carvalho, “Travel arXiv preprint arXiv:1210.2294, 2012.
time prediction using machine learning,” in Proceedings of the 4th [18] R. D. Hart and L. Osborne, “Baselining current road weather informa-
ACM SIGSPATIAL International Workshop on Computational Trans- tion: Results of the 2010 quality and importance survey,” NTL, TRIS,
portation Science. ACM, 2011, pp. 34–38. USDOT, Tech. Rep., 2011.
[7] L. Huang and M. Barth, “A novel loglinear model for freeway travel
[19] S. E. Boselly III, G. S. Doore, J. Thornes, C. Ulberg, and D. Ernst,
time prediction,” in Intelligent Transportation Systems, 2008. ITSC
“Road weather information systems volume 1: Research report,”
2008. 11th International IEEE Conference on. IEEE, 2008, pp. 210–
Strategic Highway Research Program Publication-SHRP-H-350, Na-
215.
tional Research Council, Washington, DC, pp. 90–93, 1993.
[8] E. Jenelius and H. N. Koutsopoulos, “Travel time estimation for urban
road networks using low frequency probe vehicle data,” Transportation [20] M. Riehm, “New methods for improving winter road maintenence,”
Research Part B: Methodological, vol. 53, pp. 64–81, 2013. Ph.D. dissertation, KTH, 2010.
[9] R. Logendran and L. Wang, “Dynamic travel time estimation using [21] K. Sigakova, “Road freight transport travel time prediction,” M.Sc.
regression trees,” Oregon Department of Transport Research Unit, Thesis, Blekinge Institute of Technology, Tech. Rep., 2012.
Tech. Rep., 2008. [22] GOOGLE, “Google earth open access software,” June 2014. [Online].
[10] I. Tsapakis, T. Cheng, and A. Bolbol, “Impact of weather conditions Available: https://earth.google.com/
on macroscopic urban travel times,” Journal of Transport Geography,
vol. 28, pp. 204–211, 2013.
[11] M. Yildirimoglu and N. Geroliminis, “Experienced travel time pre-
diction for congested freeways,” Transportation Research Part B: ACKNOWLEDGMENT
Methodological, vol. 53, pp. 45–63, 2013.
[12] W. Qiao, A. Haghani, and M. Hamedi, “Short-term travel time pre-
diction considering the effects of weather,” Transportation Research
We would like to express our gratitude to the Swedish
Record: Journal of the Transportation Research Board, vol. 2308, transport operator who participated in our experiment. We
no. 1, pp. 61–72, 2012. equally wish to thank the project East West Transport Cor-
[13] J. Miller, S.-i. Kim, M. Ali, and T. Menard, “Determining time to
traverse road sections based on mapping discrete gps vehicle data to
ridor II, for providing funding to carry out this research.
continuous flows,” in Intelligent Vehicles Symposium (IV), 2010 IEEE. Without their support, we would not have been able to
IEEE, 2010, pp. 615–620. conduct this research.

View publication stats