Collective traffic forecasting

Marco Lippi, Matteo Bertini, Paolo Frasconi

Dipartimento Sistemi e Informatica, Università degli Studi di Firenze

lippi@dsi.unifi.it,bertinim@dsi.unifi.it,p-f@dsi.unifi.it

Abstract. Traffic forecasting has recently become a crucial task in the

area of intelligent transportation systems, and in particular in the devel-
opment of traffic management and control. We focus on the simultaneous
prediction of the congestion state at multiple lead times and at multiple
nodes of a transport network, given historical and recent information.
This is a highly relational task along the spatial and the temporal di-
mensions and we advocate the application of statistical relational learn-
ing techniques. We formulate the task in the supervised learning from
interpretations setting and use Markov logic networks with grounding-
specific weights to perform collective classification. Experimental results
on data obtained from the California Freeway Performance Measurement
System (PeMS) show the advantages of the proposed solution, with re-
spect to propositional classifiers. In particular, we obtained significant
performance improvement at larger time leads.

1 Introduction

Intelligent Transportation Systems (ITSs) are widespread in many densely ur-

banized areas, as they give the opportunity to better analyze and manage the
growing amount of traffic flows, due to increased motorization, urbanization,
population growth, and changes in population density.
One of the main targets of an ITS is to reduce congestion times, as they
seriously affect the efficiency of a transportation infrastructure, usually measured
as a multi-objective function taking into account several aspects of a traffic
control system, like travel time, air pollution, and fuel consumption. As for
travel time, for example, it is often important to minimize both the mean value
and its variability [13], which represents an added cost for a traveler making a
given journey.
This management effort is supported by the growing amount of data gathered
by ITSs, coming from a variety of different sources. Loop detectors are the most
commonly used vehicle detectors for freeway traffic monitoring, which can typi-
cally register the number of vehicles passed in a certain time interval (flow), and
the percentage of time the sensor is occupied per interval (occupancy). In recent
years, there has been also a spread of employment of wireless sensors, like GPS
and floating car data (FCD) [11], which will eventually reveal in real-time the
position of almost every vehicle, by collecting information from mobile phones in
vehicles that are being driven. These different kinds of data are heterogeneous,
and therefore would need a pre-processing phase in order to be integrated and
used as support to the decision processes. A large sensor network corresponds to
a large number of potentially noisy or faulty components. In particular, in the
case of traffic detectors, several different fault typologies might affect the system:
communication problems on the line, intermittent faults resulting in insufficient
or incomplete data transmitted by the sensors, broken controllers, bad wiring,
etc.
In Urban Traffic Control (UTC) systems, such as the Split Cycle Offset Op-
timization Technique (SCOOT) system [15] and the Sydney Coordinated Adap-
tive Traffic (SCAT) system [16], short-term forecasting modules are used to
adapt system variables and maintain optimal performances. Systems without
a forecasting module can only operate in a reactive manner, after some event
has occurred. Classic short-term forecasting approaches usually focus on 10-15
minutes ahead predictions [19, 24, 20]. Effective proactive transportation man-
agement (e.g. car navigation systems), arguably needs forecasts extending on
longer horizons in order to be effective.
Most of the predictors employed in these traffic control systems are based
on a time series forecasting technology. Time series forecasting is a vast area of
statistics, with a wide range of application domains [3]. Given the history of past
events sampled at certain time intervals, the goal is to predict the continuation of
the series. Formally, given a time series X = {x1 , . . . , xt } describing the dynamic
behavior of some observed physical quantity xj , the task is to predict xt+1 . In
the traffic management domain, common physical quantities of interest are (i)
the traffic flow of cars passing at a given location in a fixed time interval, (ii)
the average speed observed at a certain location, (iii) the average time needed
to travel between two locations.
Historically, many statistical methods have been developed to address the
problem of traffic forecasting: these include methods based on auto-regression
and moving average, such as ARMA, ARIMA, SARIMA and other variants,
or non-parametric regression. See [19] and references therein for an overview
of these statistical methodologies. Also from a machine learning perspective,
the problem of traffic forecasting has been addressed using a wide number of
different algorithms, like support vector regression (SVR) [24], Bayesian net-
works [20] or time-delay neural networks (TDNNs) [1]. Most of these methods
address the problem as single-point forecasting, intended as the ability to predict
future values of a certain physical quantity at a certain location, given only past
measurements of the same quantity at the same location. Yet, given a graph
representing a transportation network, predicting the traffic conditions at multi-
ple nodes and at multiple temporal steps ahead is an inherently relational task,
both in the spatial and in the temporal dimension: for example, at time t, the
predictions for two measurement sites s1 and s2 , which are spatially close in
the network, can be strongly interrelated, as well as predictions at t and t + 1
for the same site s. Inter-dependencies between different time series are usually
referred to as Granger’s causality [8], a concept initially introduced in the do-
main of economy and marketing: time series A is said to Granger-cause time
series B if A can be used to enhance the forecasts on B. Few methods until now
have taken into account the relational structure of the data: multiple Kalman
filters [23], the STARIMA model (space-time ARIMA) [10] and structural time
series models [7] are the first attempts in this direction.
The use of a statistical relational learning (SRL) framework for this kind
of task might be crucial in order to improve predictive accuracy. First of all,
SRL allows to represent the domain in terms of logical predicates and rules,
and therefore to easily include background knowledge in the model, and to de-
scribe relations and dependencies, such as the topological characteristics of a
transportation network. Within this setting, the capability of SRL models to
integrate multiple sources and levels of information might become a key feature
for future transportation control systems. Moreover, the SRL framework allows
to perform collective classification or regression, by jointly predicting traffic con-
ditions in the whole network in a single inference process: in this way, a single
model can represent a wide set of locations, while propositional methods should
typically train a different predictor for each node in the graph.
Dealing with large data sets within SRL is a problem which still has to receive
adequate attention, but it is one of the key challenges of the whole research
area [5]. Traffic forecasting is a very interesting benchmark from this point of
view: for example, just considering highways in California, over 30,000 detectors
continuously generate flow and occupancy data, producing a huge amount of
information. Testing the scalability of inference algorithms on such a large model
is a crucial point for SRL methodologies.
Moreover, many of the classic time series approaches like ARIMA, SARIMA
and most of their variants, are basically linear models. Non-linearity, on the
other hand, is a crucial issue in many application domains in order to build
a competitive predictor: for this reason, some attempts to extend statistical
approaches towards non-linear models have been proposed, as in the KARIMA
or VARMA models [22, 4].
Among the many SRL methodologies that have been proposed in recent
years, we employ Markov logic [6], extended with grounding-specific weights
(GS-MLNs) [12]. The first-order logic formalism allows to incorporate back-
ground knowledge of the domain in a straightforward way. The use of probabili-
ties within such a model allow us to handle noise to take into account statistical
interdependencies. The grounding-specific weights extension enables the use of
vectors of continuous features and non-linear classifiers (like neural networks)
within the model.

2 Grounding-Specific Markov Logic Networks

Markov logic [6] integrates first-order logic with probabilistic graphical models,
providing a formalism which allows us to describe a domain in terms of logic
predicates and probabilistic formulae. While a first-order knowledge base can
be seen as a set of hard constraints over possible worlds (or Herbrand interpre-
tations), where a world violating even a single formula has zero probability, in
Markov logic such a world would be less probable, but not impossible. Formally,
a Markov logic network (MLN) is defined by a set of first-order logic formulae
F = {F1 , . . . , Fn } and a set of constants C = {C1 , . . . , Ck }. A Markov random
field is then created by introducing a binary node for each possible ground atom
and an edge between two nodes if the corresponding atoms appear together in a
ground formula. Uncertainty is handled by attaching a real-valued weight wj to
each formula Fj : the higher the weight, the lower the probability of a world violat-
ing that formula, others things being equal. In the discriminative setting, MLNs
essentially define a template for arbitrary (non linear-chain) conditional random
fields that would be hard to specify and maintain if hand-coded. The language
of first-order logic, in fact, allows to describe relations and inter-dependencies
between the different domain objects in a straightforward way. In this paper,
we are interested in the supervised learning setting. In Markov logic, the usual
distinction between the input and output portions of the data is reflected in the
distinction between evidence and query atoms. In this setting, an MLN defines
a conditional probability distribution of query atoms Y given evidence atoms
X, expressed as a log-linear model in a feature space described by all possible
groundings of each formula:
P

exp Fi ∈FY wi ni (x, y)
P (Y = y|X = x) = (1)
Zx

where FY is the set of clauses involving query atoms and ni (x, y) is the number
of groundings of formula Fi satisfied in world (x, y). Note that the feature space
jointly involves X and Y as in other approached to structured output learning.
MAP inference in this setting allows us to collectively predict the truth value
of all query ground atoms: f (x) = y ∗ = arg maxy P (Y = y|X = x). Solving the
MAP inference problem is known to be intractable but even if we could solve
it exactly, the prediction function f is still linear in the feature space induced
by the logic formulae. Hence, a crucial ingredient for obtaining an expressive
model (which often means an accurate model) is the ability of tailoring the fea-
ture space to the problem at hand. For some problems, this space needs to be
high-dimensional. For example, it is well known that linear chain conditional
random fields (which we can see as a special case of discriminative MLNs), often
work better in practice when using high-dimensional feature spaces. However,
the logic language behind MLNs only offers a limited ability for controlling the
size of the feature space. We will explain this using the following example. Sup-
pose we have a certain query predicate of interest, Query(t, s) (where, e.g., the
variable t and s represent time and space) that we know to be predictable from
a certain set of attributes, one for each (t, s) pair, represented by the evidence
predicate Attributes(t, s, a1 , a2 , . . . , an ). Also, suppose that performance for
this hypothetical problem crucially depends, for each t and s, on our ability
of defining a nonlinear mapping between the attributes and the query. To fix
our ideas, imagine that an SVM with RBF kernel taking a1 , a2 , . . . , an as inputs
(treating each (s, t) pair as an independent example) already produces a good
classifier, while a linear classifier fails. Finally, suppose we have some available
background knowledge, which might help us to write formulae introducing sta-
tistical interdependencies between different query ground atoms (at different t
and s), thus giving us a potential advantage in using a non-iid classifier for this
problem. An MLN would be a good candidate for solving such a problem, but
emulating the already good feature space induced by the RBF kernel may be
tricky. One possibility for producing a very high dimensional feature space is to
define a feature for each possible configuration of the attributes. This can be
achieved by writing several ground formulae with different associated weights.
For this purpose, in the Alchemy system1 , one might write an expression like

Attributes(t, s, +a1 , +a2 , . . . , +an ) ⇒ Query(t, s)

where the + symbol preceding some of the variables expands the expression into
separate formulae resulting from the possible combination of constants from
those variables. Different weights are attached to each formula in the resulting
expansion. Yet, this solution presents two main limitations: first, the number
of parameters of the MLN grows exponentially with the number of variables in
the formula; second, if some of the attributes ai are continuous, they need to be
discretized in order to be used within the model.
GS-MLNs [12] allow us to use weights that depend on the specific grounding
of a formula, even if the number of possible groundings can in principle grow
exponentially or can be unbound in the case of real-valued constants. Under this
model, we can write formulae of the kind:

Attributes(t, s, $v) ⇒ Query(t, s)

where v has the type of an n-dimensional real vector, and the $ symbol indi-
cates that the weight of the formula is a parameterized function of the specific
constant substituted for the variable v. In our approach, the function is realized
by a discriminative classifier, such as a neural network with adjustable parame-
ters θ. The idea of integrating non-linear classifiers like neural networks within
conditional random fields has been also recently proposed in conditional neural
fields [14].
In MLN with grounding-specific weights, the conditional probability of query
atoms given evidence can therefore be rewritten as follows:
P P 
exp Fi ∈FY j wi (cij , θi )nij (x, y)
P (Y = y|X = x) = (2)
Zx

where wi (cij , θi ) is a function of some constants depending on the specific ground-

ing, indicated by cij , and of a set of parameters θi .
Any inference algorithm for standard MLNs can be applied with no changes.
During the parameter learning phase, on the other hand, MLN and neural net-
work weights need to be adjusted jointly. The resulting algorithm can implement
1
http://alchemy.cs.washington.edu
gradient ascent, exploiting the chain rule:
∂P (y|x) ∂P (y|x) ∂wi
=
∂θk ∂wi ∂θk
where the first term is computed by MLN inference and the second term is
computed by backpropagation. As in standard MLNs, the computation of the
first term requires to compute the expected counts Ew [ni (x, y)]:

∂P (y|x) X
= ni (x, y) − P (y 0 |x)ni (x, y ∗ ) = ni (x, y) − Ew [ni (x, y)]
∂wi 0 y

which are usually approximated with the counts in the MAP state y ∗ :
∂P (y|x)
' ni (x, y) − ni (x, y ∗ )
∂wi
From the above equation, we see that if all the groundings of formula Fj are
correctly assigned their truth values in the MAP state y ∗ , then that formula gives
a zero contribution to the gradient, because nj (x, y) = nj (x, y ∗ ). For grounding-
specific formulae, each grounding corresponds to a different example for the
neural network: therefore, there will be no backpropagation term for a given
example if the truth value of the corresponding atom has been correctly assigned
by the collective inference.
When learning from many independent interpretations, it is possible to split
the data set into minibatches and apply stochastic gradient descent [2]. Basically
this means that gradients of the likelihood are only computed for small batches
of interpretations and weights (both for the MLN and for the neural networks)
are updated immediately, before working with the subsequent interpretations.
Stochastic gradient descent can be more generally applied to minibatches con-
sisting of the connected components of the Markov random field generated by
the MLN. This trick is inspired by a common practice when training neural
networks and can very significantly speedup training time.

3 Data preparation and experimental setting

3.1 The data set
We performed our experiments on the California Freeway Performance Measure-
ment System (PeMS) data set [21], which is a wide collection of measurements
obtained by over 30,000 sensors and detectors placed around nine districts in
California. The system covers 164 Freeways, including a total number of 6,328
mainline Vehicle Detector Stations and 3,470 Ramp Detectors.
The loop detectors used within the PeMS are frequently deployed as single
detectors, one loop per lane per detector station. The raw single loop signal is
noisy and can be used directly to obtain only the raw count (traffic flow) and
the occupancy (lapse of time the loop detector is active) but cannot measure the
Fig. 1. The case study used in the experiments: 7 measurement stations placed on
three different Highways in the area of East Los Angeles.

speed of the vehicles. The PeMS infrastructure collects filtered and aggregated
flow and occupancy from single loop detectors, and provides an estimate of the
speed [9] and other derived quantities. In some locations, a double loop detector
is used to directly measure the instantaneous speed of the vehicles. All traffic
detectors report measurements every 30 seconds.
In our experiments, the goal is to predict whether the average speed at a
certain time in the future falls under a certain threshold. This is the measurement
employed by GoogleTM Maps2 for the coloring scheme encoding the different
levels of traffic congestions: the yellow code, for example, means that the average
speed is below 50 mph, which is the threshold adopted in all our experiments.

Table 1. Summary of stations used in experiments.

Station VDS Highway Lanes

A 716091 I10-W 4
B 717055 I10-W 4
C 717119 I10-W 4
D 717154 I10-W 5
E 717169 I10-W 4
F 717951 I605-S 4
G 718018 I710-S 3

In our case study, we focused on seven locations in the area of East Los
Angeles (see Figure 1), five of which are placed on the I10 Highway (direction
West), one on the I5 (direction South) and one on the I710 (direction South)
(see Table 1). We aggregated the available raw data into 15-minutes samples,
averaging the measurements taken on the different lanes. In all our experiments
we used the previous three hours of measurements as the input portion of the
data. For all considered locations we predict traffic congestions at the next four
lead times (i.e., 15, 30, 45 and 60 minutes ahead). Thus each interpretation spans
a time interval of four hours. We used two months of data (Jan-Feb 2008) as
2
http://maps.google.com
 Flow Direction

E F D C A G B

E 0.96

0.88
F
0.80
D
0.72
C
0.64
A
0.56

G
0.48

B
0.40

Fig. 2. Spatiotemporal correlations in the training set data. There are 28 boolean con-
gestion variables corresponding to 7 measurement stations and 4 lead times. Rows and
columns are lexicographically sorted on the station-lead time pair. With the exception
of station E, spatial correlations among nearby stations are very strong and we can
observe the spatiotemporal propagation of the congestion state along the direction of
flow (traffic is westbound).

training set, one month (Mar 2008) as tuning set, and two months (Apr-May
2008) for test. Time intervals of four hours containing missing measurements
due to temporary faults in the sensors were discarded from the data set.
The inter-dependencies between nodes which are close in the transportation
network are evident from the simple correlation diagram shown in Figure 2.

3.2 Experimental setup

The GS-MLN model was trained under the learning from interpretations setting.
An interpretation in this case corresponds to a typical forecasting session, where
at time t we want to forecast the congestion state of the network at future
lead times, given previous measurements. Hence interpretations are indexed by
their time stamp t, which is therefore be omitted in all formulae (the temporal
index h in the formulae below refers to the time lead of the prediction, i.e. 1,2,3,
and 4 for 15,30,45, and 60 minutes ahead). Interpretations are assumed to be
independent, and this essentially follows the setting of other supervised learning
approaches such as [24, 18, 17]. However, in our approach congestion states at
different lead times and at different sites are predicted collectively. Dependencies
are introduced by spatiotemporal neighborhood rules, such as

Congestion(+s, h) ∧ Weekday(+wd) ∧ TimeSlot(+ts)

(3)
⇒ Congestion(+s, h + 1)

Congestion(+s1, h) ∧ Next(s1, s2) ⇒ Congestion(+s2, h + 1) (4)

where Congestion(S, H) is true of the velocity at site S and lead time H falls
below the 50mph threshold, and the + symbol before a site variable assigns a
different weight to each site or site pair. The predicate Next(s1, s2) is true if site
s2 follows site s1 in the flow direction. The predicate Weekday(wd) distinguishes
between workdays and holidays, while TimeSlot(ts) encodes the part of the day
(morning, afternoon, etc.) of the current timestamp.
Of course the road congestion state also depends on previously observed
velocity or flow. Indeed, literature results [24, 18, 17] suggest that good local
forecasts can be obtained as a nonlinear function of the recent sequence of ob-
served traffic flow or speed. Using GS-MLNs, continuous attributes describing
the observed time series can be introduced within the model, using a set of
grounding-specific formulae, e.g.:

SpeedSeries(SD, $SeriesD) ⇒ Congestion(SD, 1) (5)

where the grounding-specific weights are computed by a neural network taking

as input a real vector associated with constant Series SD (being SD the station
identifier), containing past speed measurements during the previous 12 time
steps. Note that a separate formula (and a separate neural network) is employed
for each site and for each lead time.
Seasonality was encoded by the predicate SeasonalCongestion(s), which is
true if, on average, station s presents a congestion at the time of the day referred
to by the current interpretation (this information was extracted from averages
on the training set). Other pieces of background knowledge were encoded in
the MLN. For example, the number of lanes at a given site can be influence
bottleneck behaviors:
Congestion(s1, h) ∧ NodeClose(s1, s2) ∧ NLanes(s1, l1) ∧ NLanes(s2, l2)∧
l2 < l1 ⇒ Congestion(s2, h + 1)

The MLN contained 14 formulae in the background knowledge and 125 param-
eters after grounding variables prefixed by a +. The 28 neural networks had 12
continuous inputs and 5 hidden units each, yielding about 2000 parameters in to-
tal. Our software implementation is a modified version of the Alchemy system to
incorporate neural network as pluggable components. Inference was performed
by MaxWalkSat algorithm. Twenty epochs of stochastic gradient ascent were
performed, with a learning rate  = 0.03 for the MLN weights, and µ = 0.00003
n
for the neural networks, being n the number of misclassifications in the cur-
rent minibatch. In order to further speed up the training procedure, all neural
networks were pre-trained for a few epochs (using the congestion state as the
target) before plugging them into the GS-MLN jointly and tuning the whole set
of parameters.
We compared the obtained results against three competitors:
Trivial predictor The seasonal average classifier predicts, for any time of the
day, the congestion state observed on average in the training set at that
time. Although it is a baseline predictor, it is widely used in literature as a
competitor.
SVM We used SVM as a representative of state-of-the-art propositional clas-
sifiers. A different SVM with RBF kernel was trained for each station and
for each lead time, performing a separated model selection for the C and γ
values to be adopted for each measurement station. The features used by
the SVM predictor consist in the speed time series observed in the past 180
minutes, aggregated at 15 minutes intervals, hence producing 12 features,
plus an additional feature representing the seasonal average at current time.
A gaussian normalization was applied to all the features.
Standard MLN When implementing the classifier based on standard MLNs,
the speed time series had to be discretized in order to be used within the
model. Five different speed classes were used, and the quantization thresholds
were chosen by following a maximum entropy strategy. The trend of the speed
time series was modeled by the following set of formulae that were used in
place of formula 5:

Speed Past 1(n, +v) ⇒ Congestion(n, 1)

···
Speed Past k(n, +v) ⇒ Congestion(n, 1)

where predicate Speed Past j(node, speed value) encodes the discrete val-
ues of the speed at the j-th time step before the current time. Note that
an MLN containing only the above formulae essentially represents a logistic
regression classifier taking the discretized features as inputs. All remaining
formulae were identical to those used in conjunction with the GS-MLN.
As for the predictor based on GS-MLNs, there is no need to use discretized
features, but the same vectors of features used by the SVM classifier can be
adopted.

4 Results and discussion

4.1 Performance analysis
The congestion state in the analyzed highway segment is a very unbalanced task
even at the 50mph threshold. Table 2 shows the percentage of positive query
atoms in the training set and in the test set, for each station. The last two
columns report the percentage of days containing at least one congestion. The
data distribution shows that the stations present different behaviors, corrobo-
rating the choice of using different neural networks for each station.

Table 2. Percentage of true ground atoms, for each measurement station. The per-
centage of days in the train/test set containing at least one congestion is reported in
the last two columns.

Station % pos train % pos test % pos days train % pos days test
A 11.8 9.2 78.3 70.7
B 5.8 4.9 60.0 53.4
C 16.8 13.7 66.6 86.9
D 3.4 2.3 45.0 31.0
E 28.2 22.9 86.7 72.4
F 3.9 1.8 51.6 31.0
G 1.9 1.7 30.0 22.4

Given the unbalanced data set, we compare the predictors on the F1 measure,
as the harmonic mean between precision P = T PT+F P TP
P and recall R = T P +F N :
2P R
F1 = P +R . Table 3 shows the F1 measure, averaged per station. The advantages
of the relational approach are much more evident when increasing the prediction
horizon: at 45 and 60 minutes ahead, the improvement of the GS-MLN model is
statistically significant, according to a Wilcoxon paired test, with p-value< 0.05.
Detailed comparisons for each sensor station at 15, 30, 45, and 60 minutes ahead
are reported in Tables 4 , Tables 5 , Tables 6 and 7, respectively. These tables
show that congestion at some of the sites are clearly “easier” to predict than at
other sites. Comparing Tables 4-7 to Table 2 we see that the difficulty strongly
correlates with the data set imbalance, an effect which is hardly surprising. It
is also often the case that GS-MLN significantly outperforms the SVM classifier
for “difficult” sites. The comparison between the standard MLN and the GS-
MLN shows that input quantization can significantly deteriorate performance,
all other things being equal. This supports the proposed strategy of embedding
neural networks as a key component of the model.
An interesting performance measure considers only those test cases in which
traffic conditions are anomalous with respect to the typical seasonal behavior.
To this aim, we restricted the test set, by collecting only those interpretations
for which the baseline seasonal average classifier would miss the prediction of
the current congestion state. Table 8 shows that the advantage of the relational
approach is still evident for long prediction horizons.
The experiments were performed on a 3GHz processor with 4Mb cache. The
total training time for SVM is 40 minutes, and 7-8 hours for GS-MLNs. As for
testing times, both systems perform in real-time.
Table 3. Comparison between the tested predictors. Results show the F1 on the posi-
tive class, averaged on the seven nodes. The symbol indicates a significant loss of the
method with respect to GS-MLN, according to a Wilcoxon paired test (p-value<0.05).

15 m 30 m 45 m 60 m
Seasonal Avg 38.3 38.3 38.3 38.3
SVM 81.7 68.6 56.4 51.8
MLN 59.5 56.5 53.6 50.4
GS-MLN 80.9 69.2 61.6 56.9

Table 4. Details on the predictions per station, at 15 minutes ahead.

SVM MLN GS-MLN

A 82.9 64.0 80.4
B 78.0 50.8 74.5
C 91.2 66.5 89.1
D 77.5 51.9 79.5
E 92.0 69.4 92.9
F 70.6 51.9 66.7
G 80.0 61.7 83.4

Table 5. Details on the predictions per station, at 30 minutes ahead.

SVM MLN GS-MLN

A 76.2 50.6 74.2
B 60.9 46.5 60.5
C 85.6 81.5 86.0
D 64.4 57.0 65.5
E 85.7 74.3 86.0
F 36.0 30.4 45.6
G 71.6 55.5 66.7

Table 6. Details on the predictions per station, at 45 minutes ahead.

SVM MLN GS-MLN

A 74.3 71.1 73.5
B 41.6 29.3 44.5
C 82.7 75.1 83.9
D 46.2 49.9 59.4
E 80.7 78.2 82.9
F 33.8 28.7 37.3
G 35.5 43.2 50.0
 Table 7. Details on the predictions per station, at 60 minutes ahead.

SVM MLN GS-MLN

A 72.5 71.6 72.1
B 29.9 27.9 37.0
C 83.5 80.9 84.7
D 38.0 41.0 52.4
E 79.7 75.4 79.9
F 26.0 21.0 29.9
G 33.3 32.6 42.4

Table 8. Comparison between the tested predictors, only on those cases where the
seasonal average predictor fails. Results show the F1 on the positive class, averaged on
the seven nodes.

15 m 30 m 45 m 60 m
SVM 81.4 69.1 59.1 59.2
MLN 39.9 47.6 48.4 41.6
GS-MLN 78.4 68.2 68.4 65.5

4.2 Dealing with missing data

The problem of missing or incomplete data is crucial in all time series forecasting
applications [3, 4]: in the case of punctual missing information, a reconstruction
algorithm might be employed in order to interpolate the signal, so that prediction
methods might be applied unchanged. Occasionally, sensor faults can last several
time steps, and when this happens, a large part of the input can be unavailable to
a standard propositional predictor until the sensor recovers from the failure state.
Of course, cases containing missing data can be filtered from the training set as
we did for our previous experiments. However, in order to deploy a predictor
on a real-time task, it is necessary also to handle the case of missing values at
prediction time. A relational model can be in principle more robust than its
propositional counterpart by exploiting information from nearby sites.
In this section we report results obtained by simulating the absence of several
values within the observed time series, using the seasonal average predictor (Sec-
tion 3.2) as reconstruction algorithm for these unobserved data . Producing an
accurate model of sensor faults is clearly beyond the scope of this paper and we
built a naive observation model based on a two states first-order Markov chain
with P (observed 7→ observed) = 0.99 and P (reconstructed 7→ reconstructed) =
0.9. The performance of the predictors on this task are shown in Table 9.

5 Conclusions

We have proposed a statistical relational learning approach to traffic forecast-

ing, in order to collectively classify the congestion state at several nodes of a
Table 9. Comparison between the tested predictors, using a test set containing missing
values, reconstructed using the seasonal average. Results show the F1 on the positive
class.

15 m 30 m 45 m 60 m
SVM 79.0 63.2 53.6 48.8
GS-MLN 80.5 70.4 62.6 58.1

transportation network, and at multiple lead times in the future, exploiting the
relational structure of the domain. Our method is based on grounding-specific
Markov logic networks, which extend the framework of Markov logic in order to
include discriminative classifiers and generic vectors of features within the model.
Experimental results performed on a case study extracted from the Californian
PeMS data set show that the relational approach outperforms the propositional
one, in particular when the prediction horizon grows.
Although we performed experiments on a binary classification task, we plan
to extend the framework also to the case of multiclass classification or ordinal
regression. As a further direction of research, the use of Markov logic gives the
possibility to extend the model by applying structure learning algorithms to
learn relations and dependencies directly from data in an automatic way.
The proposed methodology is not restricted to traffic management, but it can
be applied to several different time series application domains, such as ecologic
time series, for air pollution monitoring, or economic time series, for marketing
analysis.

Acknowledgments
This research is partially supported by grant SSAMM-2009 from the Foundation
for Research and Innovation of the University of Florence.

References
1. B. Abdulhai, H. Porwal, and W. Recker. Short-term freeway traffic flow predic-
tion using genetically optimized time-delay-based neural networks. Transportation
Research Board, 78th Annual Meeting, Washington D.C, 1999.
2. Léon Bottou. Stochastic learning. In Olivier Bousquet and Ulrike von Luxburg,
editors, Advanced Lectures on Machine Learning, Lecture Notes in Artificial Intel-
ligence, LNAI 3176, pages 146–168. Springer Verlag, Berlin, 2004.
3. G. Box, G. M. Jenkins, and G. Reinsel. Time Series Analysis: Forecasting &
Control (3rd Edition). Prentice Hall, 3rd edition, February 1994.
4. C. Chatfield. The Analysis of Time Series: An Introduction. Chapman &
Hall/CRC, sixth edition, July 2003.
5. T. G. Dietterich, P. Domingos, L. Getoor, S. Muggleton, and P. Tadepalli. Struc-
tured machine learning: the next ten years. Machine Learning, 73(1):3–23, 2008.
6. P. Domingos, S. Kok, D. Lowd, H. Poon, M. Richardson, and P. Singla. Markov
logic. In Probabilistic Inductive Logic Programming, pages 92–117, 2008.
 7. B. Ghosh, B. Basu, and M. O’Mahony. Multivariate short-term traffic flow fore-
casting using time-series analysis. Trans. Intell. Transport. Sys., 10(2):246–254,
2009.
8. C. W. J. Granger and Paul Newbold. Forecasting Economic Time Series (Economic
Theory and Mathematical Economics). Academic Press, 1977.
9. Z. Jia, C. Chen, B. Coifman, and P. Varaiya. The pems algorithms for accurate,
real-time estimates of g-factors and speeds from single-loop detectors. pages 536
–541, 2001.
10. Y. Kamarianakis and P. Prastacos. Space-time modeling of traffic flow. Comput.
Geosci., 31:119–133, March 2005.
11. B. S. Kerner, C. Demir, R. G. Herrtwich, S. L. Klenov, H. Rehborn, M. Aleksic,
and A. Haug. Traffic state detection with floating car data in road networks. In
Intelligent Transportation Systems, 2005. Proceedings. 2005 IEEE, pages 44–49,
2005.
12. M. Lippi and P. Frasconi. Prediction of protein beta-residue contacts by markov
logic networks with grounding-specific weights. Bioinformatics, 25(18):2326–2333,
2009.
13. R. B. Noland and J. W. Polak. Travel time variability: a review of theoretical and
empirical issues. Transport Reviews: A Transnational Transdisciplinary Journal,
22:39–54, 2002.
14. J. Peng, L. Bo, and J. Xu. Conditional neural fields. In Y. Bengio, D. Schuur-
mans, J. Lafferty, C. K. I. Williams, and A. Culotta, editors, Advances in Neural
Information Processing Systems 22, pages 1419–1427. 2009.
15. D. L. Selby and R. Powell. Urban traffic control system incorporating scoot: design
and implementation. Institution of Civil Engineers Proceedings, 82:903–20, Oct
1987.
16. AG. Sims. S.c.a.t. the sydney co-ordinated adaptive traffic system. Symposium on
Computer Control of Transport 1981: Preprints of Papers, pages 22–26, 1981.
17. B. L. Smith and M. J. Demetsky. Short-term traffic flow prediction: neural network
approach. Transportation Research Record, 1453:98–104, 1997.
18. B. L. Smith and M. J. Demetsky. Traffic flow forecasting: Comparison of modeling
approaches. Journal of Transportation Engineering-Asce, 123(4):261–266, Jul-Aug
1997.
19. B. L. Smith, B.M. Williams, and R. Keith Oswald. Comparison of parametric and
nonparametric models for traffic flow forecasting. Transportation Research Part C,
10(4):303–321, 2002.
20. S. Sun, C. Zhang, and G. Yu. A bayesian network approach to traffic flow fore-
casting. IEEE Transactions on Intelligent Transportation Systems, 7(1):124–132,
2006.
21. P. Varaiya. Freeway Performance Measurement System: Final Report. PATH
Working Paper UCB-ITS-PWP-2001-1, University of California Berkley, 2001.
22. S. Watson. Combining kohonen maps with arima time series models to forecast traf-
fic flow. Transportation Research Part C: Emerging Technologies, 4:307–318(12),
October 1996.
23. J. Whittaker, S. Garside, and K. Lindveld. Tracking and predicting a network
traffic process. International Journal of Forecasting, 13(1):51–61, 1997.
24. C. H. Wu, J. M. Ho, and D. T. Lee. Travel-time prediction with support vector
regression. Ieee Transactions On Intelligent Transportation Systems, 5(4):276–281,
December 2004.