2015 12th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD)

Fuzzy c-Means Clustering Identification Method of

Urban Road Traffic State
Guangyu Zhu , Jianjun Chen Peng Zhang
MOE Key Laboratory for Transportation Complex Systems Theory Beijing Transportation Research Center
and Technology Beijing China 100044
Beijing Jiaotong University
Beijing China 100044

Abstract—Urban road state identification refers to determining traffic flow parameters. All the above studies have provided
the operation status of the road network system, which plays an premises and laid a solid foundation for this paper.
important role in urban road traffic management. By clustering
time series of traffic flow, typical fluctuation pattern recognize As road traffic system is a complex system, traffic flow
algorithms of traffic flow can get the urban road network parameters will change along with time randomly due to
operation states. As the detected traffic data contain vague and impacts of some factors, such as the level of urban economics,
uncertain information, preprocessing is needed. An improved travel modes, driving habits, weather and other environmental
fuzzy c-means clustering (FCM) method is proposed in this factors, which may incur errors during data detection. The
paper. A case study based on urban road section of Beijing City uncertainty of traffic flow data directly results in the fuzziness
demonstrates the feasibility and effectiveness of the improved
FCM algorithm.
and uncertainty in representing traffic state. The boundary
between certain two traffic flow states cannot be split clearly.
Keywords-Urban Road, Traffic State Identification, Clustering, Thus people can only approximate describe the traffic state
Fuzzy c-Means with some fuzzy concepts, such as congestion, slowing down
and unblocked, etc. The key part of AID algorithm is to
I. INTRODUCTION distinguish each state from state-set by clustering the traffic
data.
Traffic state refers to the operation status of urban road
traffic network, and traffic state identification means After the fuzzy theory introduced by Zadeh, the researchers
recognizing the traffic state by using mathematical models. put the fuzzy theory into clustering. Fuzzy algorithms can
assign data object partially to multiple clusters. The degree of
Along with the increasing of the urban size as well as the membership in the fuzzy clusters depends on the closeness of
total amount of motor vehicles, traffic congestion and energy the data object to the cluster centers [7]. In the experimental
consumption are becoming a normal state. Trustworthy traffic data processing, fuzzy clustering theory has the advantage over
information, as well as most appropriate traffic-control traditional statistic theory, namely that its results sufficiently
strategies, can only be made on the basic of accurate judgment embody the inherent properties of a system with less
of urban road traffic state. The research of urban road traffic experimental data and unknown systemic probability [8]. So
state identification methodology plays an important role in the the method based on fuzzy processing technology began to
practice of managing urban traffic flow. play a more and more important role in automatic road traffic
Through years of research and practice, scholars have incident identification [9~14].
developed various methods to identify urban road traffic state. Fuzzy c-means (FCM) algorithm is a clustering algorithm
Automatic traffic state detection algorithm, also known as the based on fuzzy divisions [15, 16]. It is designed to maximize
Automatic Incident Detection (AID) algorithm was first the similarity between objects in the same cluster while
developed by California Department of Transportation in the minimizing the similarity between objects in different clusters.
1960s.It was modeled based on the fact that when an event FCM usually describes the clustering problem as a constrained
occurs, occupancy will increase in the upstream detection nonlinear programming problem, and gets the optimal fuzzy
section and will decrease in the downstream detection section. division and clustering results using the iterative optimization
In addition, flow volume, speed, and other parameters detected theory. In addition, a FCM is used because it requires smaller
with detectors are also the basis of AID algorithm research. training sets and shorter training times. Moreover, it is
The research contents and methods of AID have changed a compatible with future research needs to compare other
lot as the road traffic systems are becoming more and more systems based on cluster variances [17]. It is also useful for
complex. Series of new AID methods based on ANN (artificial specify number and shape of membership functions which
neural network) [1~4], wavelet analysis [5], SVM(Support consider from the distribution of data points [18]. Thus, it is an
Vector Machine)[6]and fuzzy theory came into being ideal algorithm for traffic state identification [19].
subsequently in 2000s. The main idea of such method is to For this reason, we adopt the fuzzy c-means (FCM)
evaluate different traffic state through clustering the data of algorithm to identify traffic state in this paper. The key of using

Authorized licensed use limited to:©2015

978-1-4673-7682-2/15/$31.00 Chang'an University. Downloaded on October
IEEE 302 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.
fuzzy clustering algorithm for traffic state identification lies in n

defining the algorithm parameters. For this purpose, two FCM ∑u

j =1
m
ij xj
(3)
parameters, i.e. the fuzzy weighting index m and the cluster Vi = n
number c, are studied. Their values are determined by the ∑u m
ij
function value of clustering effectiveness, which helps make j =1

the results more effective. Additionally, we use the cross and

assessment on the false rate to evaluate the FCM algorithm.
1 (4)
The rest contents of this paper are arranged as following: u ij = 2 /( m −1)
c ⎛ d ij ⎞
1) An improved FCM algorithm is proposed in part Ⅱ. ∑ ⎜⎜ d
k =1 ⎝ kj ⎠
⎟
⎟
2) A case is studied to validate the improved FCM model
on a real-world dataset in part Ⅲ. The procedure of FCM algorithm is composed with the
following steps:
II. IMPROVED FCM CLUSTERING ALGORITHME
Step 1: Define the cluster c and the fuzzy weighting index m,
A. Introduction to FCM Algorithm
set the iteration termination threshold ε, the initial clustering
Assume the clustering objective X = {x1 , x2 , …, xn } ⊂ R s is a
s center V (0)
, and set the iterations 0 to r.
finite data-set sample from the s-dimensional data-space R ,
where n is the number of figures to be classified in the sample Step 2: Calculate the membership matrix U (r )
with the
data-set, as is the number of clusters contained in the data-set.
following formula:
The fuzzy division is represented with the membership matrix
U = [ μij ]c×n where μij in the matrix U refers to the membership ⎧ 1
dij( r ) > 0
⎪ c 2/( m −1)
of the data point j ( j = 1, 2,......, n) relative to the cluster ⎪ ⎛ d (r )
⎞
∑ ⎜
ij
⎪ k =1 ⎜ d kj ⎟
( r ) ⎟ (5)
i(i = 1, 2,......, c) . μij is supposed to satisfy the following ⎪ ⎝ ⎠
uij( r ) =⎨
conditions: ⎪1 d ij( r ) = 0且i = j
⎪
⎧ ⎪
⎪0 ≤ u ≤ 1, ∀i = 1,..., c; ∀j = 1,..., n (1) ⎪ d ij( r ) = 0且i ≠ j
⎪ ij
⎩0
⎪ c
⎨∑ uij = 1, ∀j = 1,..., n ( r +1)
⎪ i =1 Step 3: Update the clustering center V with the
⎪ n
⎪0 ≤ ∑ uij ≤ n, ∀i = 1,..., c following formula:
⎩ j =1

n
The objective function of FCM algorithm is: ∑ (u
j =1
(r ) m
ij ) xj
(6)
( r +1)
c n Vi = n
J m (U , V1 ,...,Vc ) = ∑∑ u d m
ij
2
ij ∑ (u (r ) m
ij )
i =1 j =1
(2) j =1

In the formula (2), J m (U ,V ) is the objective function, Step 4: In the case of V ( r +1) −V ( r ) ≤ ε , terminate the

V = [V1 , ..., Vc ] is the clustering center, m ∈ [1, +∞ ) is the algorithm and export the membership matrix U and the
fuzzy weighting index, and d ij = Vi − x j is the distance clustering center V; otherwise, set r ⇐ r + 1 and go back to

between the j-th data and the i-th clustering center. Step 2 until meet the termination condition of the algorithm.

An iterative procedure is used to get the membership matrix U For the preceding algorithm, it is acceptable to initialize the

and the clustering center V when the conditions are satisfied, membership matrix first, and then perform the iteration

while minimizing the objective function J m (U , V ) . Formula (2) process.

can be solved with Lagrange method of multipliers. Shown as B. Research on m and c Optimization
(3) and (4). In this section, we introduce an effective criterion to select
the optimal parameters of the FCM. Clustering algorithm is
designed to classify the given data sample set, to maximize the
distance between clusters while minimizing the distance
between data samples within a cluster. The indicator Xie-Beni

Authorized licensed use limited to: Chang'an University. Downloaded on October

303 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.
is put forward based on this idea. The validity function of the A. Pretreatment of the Road Section Traffic Data
cluster Xie-Beni is shown as the following formula (7): As the traffic data for the selected road section was
c n collected by a detector every two minutes, there are about 450
(∑∑ (uij ) m x j − vi ) / n
2
data points contained in the data-set collected between 6:00
v XB = i =1 j =1 (7) a.m. and 9:00 p.m. And as the data was uploaded at certain
2
min vi − v j time with special interval, the uploaded data is used to
i≠ j
represent the data during a given period of time [20]. For
In the formula (7), the numerator indicates the average sum example, the data detected at 10:00 a.m. represents the traffic
of deviation between samples and different clustering centers. condition between 10:00 a.m. and 10:02 a.m.
The denominator indicates the minimum distance between
Table 1. Traffic flow parameters and their value detected on one road section
clustering centers. The smaller the numerator is, the higher the
tightness is. In contrast, the bigger the denominator is, the
Traffic flow Parameters and their values
higher the clustering separation is. As a result, a smaller vXB Time
means better clustering. Traffic Volume Speed Occupancy

So, the formula (7) is used as the criteria for the selection of 6:00 16.29 31.97 2.00
m and c. In addition, we improve the above-stated FCM 6:02 18.23 37.78 2.26
algorithm as follows: 6:04 18.69 36.35 3.55
Step 1: Define the initial values of the cluster c and the 6:06 20.21 39.54 4.14
fuzzy weighting index m, which are c0 and m0 , respectively, 6:08 21.27 54.15 4.83
and set k to 0; and define the iteration termination threshold … … … …
ε and the initial clustering center V (0) , and set r to 0. 13:22 38.50 38.59 21.71
13:24 37.97 37.70 21.43
(r )
Step 2: Calculate the membership matrix U based on 13:26 39.93 32.89 23.43
the formula (5).
13:28 41.38 31.97 24.72
( r +1)
Step 3: Update the clustering center V based on the 13:30 40.66 22.79 24.65
formula (6). … … … …
Step 4: In the case of V ( r + 1) −V ( r ) ≤ ε , export the 20:50 30.15 57.79 13.25
membership matrix U k and the clustering center V k , 20:52 28.60 60.02 11.58
otherwise, set r to make it equal ( r + 1) and go back to Step 2. 20:54 27.97 59.63 10.63
20:56 28.58 68.58 10.83
Step 5: Calculate the value of clustering validate based on the
20:58 26.61 59.70 8.03
formula (7), which is v XBk . In the case of v XBk ≥ v XBk −1 ,
terminate the algorithm and export the membership matrix
U k − 1 and the clustering center V k −1 ; otherwise, set B. FCM Clustering and Result Analysis
k ⇐ k + 1 , choose the new values of the cluster c, ck and the (1) Definition of the values of m and c
fuzzy weighting index mk , respectively, and go to Step 1. First, it is necessary to define the value range of the fuzzy
weighting index m and the cluster c.
III. CASE STUDY According to existing research, the value range of m is (1,
In this part, we perform the case study in order to validate 2.5]. With the heuristic method, we find the value of m from
the effectiveness of the improved FCM algorithm. A section of the range (1, 2.5] with an increment of 0.3, that is, m ∈ {1.3,
the freeway from the South 4th Ring Road in Beijing city, 1.6, 1.9, 2.2, 2.5}.
which consists of four lanes, is selected as research subjects. A For the value of c, the rule cmax ≤ 2 ln n is followed. In this
piece of data, on June 30, 2011, from 6:00 a.m. and 9:00 p.m.,
paper, there are 450 traffic data samples with
which is recorded every two minutes, is get from traffic flow
cmax ≤ 2ln n = 2ln 450 = 12.22 . As a result, c is a positive
database and utilized to test the algorithm. We firstly pre-treat
the collected data according to the above algorithm, and then integer which value is between 2 and 12.
apply the improved FCM algorithm for clustering. The result Second, it is necessary to calculate the clustering validity
will be analyzed at last. Finally, complete content and functions vXB (m, c) for different m and c values, to determine
organizational editing before formatting. the ideal m and c values. Set the iteration termination threshold

Authorized licensed use limited to: Chang'an University. Downloaded on October

304 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.
ε to “1×10-3. Table 2. shows the values of clustering validity According to Table 2, the clustering validity function v XB ( m, c)
functions vXB (m, c) for different m and c values. gets the minimum value in the case of m = 2.5 and c = 5 ,
Table 2. v ( m, c ) for different m and c values
which are then the ideal values of m and c.
XB
(2) Results of Clustering
Weighting According to the analysis in the section 2.2, the ideal value
index (m) m=1.3 m=1.6 m=1.9 m=2.2 m=2.5 of traffic data clustering c is 5. In other words, traffic state is
Cluster (c) ideally divided into five classes, including Very Smooth,
Smooth, Mildly Congested, Moderately Congested, and
c=2 0.2977 0.2804 0.2349 0.2101 0.1854 Congested. By using the FCM algorithm for clustering and
c=3 0.2765 0.2592 0.2320 0.1989 0.1660
setting the iteration termination threshold ε to “1×10-3,” we can
get the following clustering centers:
c=4 0.2569 0.2018 0.1852 0.1528 0.1145
⎡ 30.4321 67.5192 11.8023 ⎤
c=5 0.2266 0.2005 0.1890 0.1531 0.1138* ⎢36.0899
⎢ 53.5195 19.7148 ⎥⎥
c=6 0.2315 0.2028 0.1579 0.1538 0.1149 V = (q, v, o) = ⎢ 41.7339 42.3633 26.1723⎥
⎢ ⎥
c=7 0.2861 0.2213 0.1887 0.1537 0.1142 ⎢ 42.4389 30.0991 28.6040 ⎥
⎢⎣ 49.8657 25.7367 39.2614 ⎥⎦
c=8 0.4150 0.2606 0.1997 0.1569 0.1815
c=9 0.2703 0.2324 0.2870 0.2354 0.1701 The 1st to 5th rows of V are the clustering centers of the
five traffic states, including Very Smooth, Smooth, Mildly
c=10 0.3134 0.3088 0.2597 0.2024 0.1484 Congested, Moderately Congested, and Congested, which are
c=11 0.2910 0.3752 0.3009 0.2217 0.1411 represented by “1”, “2”, “3”, “4” and “5” respectively. The
c=12 0.3822 0.3246 0.2322 0.1585 0.1205
three data points in each row shows traffic volume, speed and
occupancy, respectively. Table 3. shows the identification
results of some points on intermediate steps.

Table 3. Traffic States of Some Selected Sample-points

No of the Traffic Corresponding

speed occupancy μ1 μ2 μ3 μ4 μ5
Smaple point volume States
25 34.51 53.59 19.73 0.0385 0.8833 0.0445 0.0207 0.0130 Smooth

127 49.05 18.41 41.66 0.0394 0.0613 0.1027 0.1878 0.6089 Congested
Mildly
161 44.58 42.21 26.92 0.0296 0.0776 0.7326 0.1077 0.0526
Congested
Moderately
331 43.85 30.25 29.25 0.0103 0.0199 0.0527 0.8635 0.0536
Congested
408 31.28 69.46 10.08 0.8683 0.0643 0.0311 0.0208 0.0156 Very Smooth

μ1 , μ 2 , μ 3 , μ 4 , μ 5 represent the membership of very Smooth, Table 4. The Clustering result of Traffic States
Smooth, Mildly Congested, Moderately Congested, and
Very Mildly Moderately
Congested, respectively. As to sample point No.25 in table 3, Traffic states Smooth Congested
its membership to smooth is 0.8833, greater than the Smooth Congested Congested
memberships to the other 4 states. So the corresponding state number of
35 105 115 117 78
of No.25 is smooth. Thus we can get the results of samples
identification on all sample points, shown as Table 4. And the
result is essentially in agreement with the real states on road. Proportion 7.78% 23.33% 25.56% 26.00% 17.33%
Fig. 2 shows the traffic state for each sample point though
clustering.

Authorized licensed use limited to: Chang'an University. Downloaded on October

305 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.
 The false rate is calculated using the cross assessment on
the false rate of the clustering results in section 4.2. Fig. 4
shows the result of the traffic state misjudgment:

Error rate
Fig. 2. Traffic state diagram

The clustering process of improved FCM is convergence.

Fig. 4. Traffic state misjudgment results
Shown as Fig. 3.
According to Fig. 4, the traffic states of 18 sample points
are inconsistent with the clustering results, obtaining a false
rate of 4 percent. In other words, the accuracy of identification
with FCM clustering algorithm is 96%, which is relatively
high. In addition, the misjudged samples changed between
Objective function×10

adjacent states without rapid changes (that is, there were no

absolute error values larger than 2). Therefore, it is feasible to
use FCM clustering algorithm to identify the road traffic state.

IV. CONCLUSIONS
In this paper, FCM algorithm is improved to identify urban
3

traffic state and empirical research is conducted based on a

real-world dataset which are actually observed against a
sectional freeway. The observed data is identified, restored and
smoothed based on empirical knowledge. Furthermore, a
clustering effectiveness function is taken as the criteria of
Fig. 3. Convergence Curve of Improved FCM Algorithm
optimizing FCM parameters. The experiment result shows the
feasibility of using the improved FCM algorithm to identify
C. Traffic State Identification Algorithm Evaluation road traffic states.
It is necessary to evaluate the algorithm after traffic state
identification, to understand the results of the identification
with the algorithm. Cross assessment on false rate [21] is a Acknowledgment
common method to evaluate the result. Using cross This work was joint-supported by “the Research Project
assessment, traffic data is clustered using FCM algorithm after for outstanding Scholars of Beijing Municipal Commission of
being processed. In addition, the category of each data point is Transport(No.kj2013-2-34)” and “the Fundamental Research
determined to eliminate each data sample in the set. Clustering Funds for the Central Universities”(No. 2014JBM211).
analysis is performed on the data set using the remaining n-1
samples in order to get the membership matrix and the
clustering center. With the new membership matrix and the REFERENCES
clustering center, the category of eliminated data points is
[1] Stephen G, Ritchie, Ruey L, Chew. Neural Network Models for
identified, which can also be compared with the original Automated Detection of Non-Recurring Congestion. Transportation
category. If the results are inconsistent, the judgment is Research Part C, 1995, 3(6):371-388.
Incorrect, then count the number of incorrectness data points http://escholarship.org/uc/item/6r89f2hw
and divide it by the total number of data points in the data set. [2] Chew.R.L, Rtehie.S.G. Automated Detection of Lane-blocking Freeway
Then, we get the false rate. A lower false rate means that the Incidents Using Artificial Neural Networks. Transportation Research
Part C, 1996, 07:261-280.
algorithm of traffic state identification works well.
http://trid.trb.org/view.aspx?id=461798

Authorized licensed use limited to: Chang'an University. Downloaded on October

306 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.
[3] WEN Juan, HE Guo-guang. Automatic Detection Algorithm of Traffic http://www.sciencedirect.com/science/article/pii/S0968090X07000034
IncidentUsing Neural Network Based on Rough Sets [13] HAO Yuan, XU Tiandong, SUN Lijun. Fuzzy Logics Based Traffic
Filtration.Communication and Transportati0n Systems Engineering and State Identification on Urban Expressway. Highway Engineering, 2008,
Information,2004,04:54-59. 33(2):94-99.
http://en.cnki.com.cn/Article_en/CJFDTotal-YSXT200404007.htm http://en.cnki.com.cn/Article_en/CJFDTOTAL-ZNGL200802023.htm
[4] WU Weitiao, JIN Wenzhou, LIN Peiqun. The Method of Traffic State [14] REN Qiliang, XIAO Yurain. Research on H-Fuzzy Evaluation Method
Identification Based on BP Neural Network. Journal of Transport for Traffic Congestion of City Road-Network . Journal of Chongqing
Information and Safety, 2011, 04:71-74+79. Jiaotong University (Natural Sciences), 2008, 25(9):2768-2770.
http://lib.cqvip.com/qk/91770A/201104/38996435.html http://ieeexplore.ieee.org/xpl/articleDetails.jsp?reload=true&arnumber=
[5] Hualiang Teng, Yi Qi.Application of wavelet technique to freeway 4737722&sortType%3Dasc_p_Sequence%26filter%3DAND(p_IS_Nu
incident detection . mber%3A4737706)
Transportation Research, 2003,11(C):289-308. [15] Kamel M S, Selim S Z. New algorithms for solving the fuzzy c-means
http://www.sciencedirect.com/science/article/pii/S0968090X03000214 Clustering problem. Pattern Recognition, 1994, 27:421.
[6] Fang Yuan, Ruey Long Cheu. Incident detection using support vector http://www.sciencedirect.com/science/article/pii/003132039490118X
machines. Transportation Research Part C, 2003, 11: 309-328. [16] Pal N R, Pal K, Keller J M, Bezdek J C. A Possibilistic Fuzzy c-Means
http://trid.trb.org/view/664196 Clustering Algorithm. 2005 13(4), 517-530.
[7] Izakian H, Abraham A, Snasel V. Fuzzy clustering using hybrid fuzzy c- http://www.researchgate.net/publication/3336300_A_Possibilistic_Fuzz
means and fuzzy particle swarm optimization. //Nature & Biologically y_c-Means_Clustering_Algorithm
Inspired Computing, 2009. NaBIC 2009. World Congress on. IEEE, [17] Wachs J.P., Stem H., Edan Y. Real-Time Hand Gesture Telerobotic
2009:1690 - 1694. System Using the Fuzzy C-Means Clustering Algorithm. World
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5393618&tag=1 Automation Congress, 2002, 403–410.
[8] Han J, Wang L, Cheng N, et al. Thermal error modeling of machine tool http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1049576
based on fuzzy c -means cluster analysis and minimal-resource [18] Somsung K. Pratishthananda S. Fuzzy supervisory control using fuzzy
allocating networks. The International Journal of Advanced c-means clustering combined with fuzzy gain scheduling. TENCON
Manufacturing Technology, 2012, 60(5-8):463-472. 2004. 2004 IEEE Region 10 Conference, 2004, 4:506-509.
http://link.springer.com/article/10.1007/s00170-011-3619-5 http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1414981
[9] Hsiao C. H. Lin C. Application of Fuzzy Logic and Neural Networks to [19] Yang Zuyuan, Huang Xiyue, Du Changhai, Tang Mingxia. Study of
Automatically Detect Freeway Traffic Incidents. Journal of urban traffic congestion judgment based on FFCM clustering.
Transportation Engineering, 1993, 125(5). Application Research of Computers, 2008, 25(9):2768-2770.
http://cedb.asce.org/cgi/WWWdisplay.cgi?90280 http://en.cnki.com.cn/Article_en/CJFDTOTAL-JSYJ200809056.htm
[10] Qingdong Da, Guowu Zhang. Quantitative Methods of Traffic [20] Pei Yulong, Ma Ji. Screening and restoration of real-time traffic data.
Congestion Analysis. Journal Of Transportation Systems Engineering Civil Engineering Journal, 2003.36(7):78-83.
And Information Technology, 2002, 2(4):45-48. http://d.wanfangdata.com.cn/Periodical_tmgcxb200307016.aspx
http://d.wanfangdata.com.cn/Periodical_jtysxtgcyxx200204009.aspx [21] Jiang Guiyan, Jiang Longhui, Zhang Xiaodong, Wang Jiangfeng.
[11] Srinivasan D., Cheu R. L., Poh Y. P. Development of an intelligent Method of Fault identification and fix based on dynamic traffic data.
technique for traffic network incident detection. Engineering Traffic and Transportation Engineering, 2004, 4(1):121-125.
Applications of Artificial Intelligence, 2002, 13(3):311-322. http://lib.cqvip.com/qk/90752A/200401/9433962.html
http://www.sciencedirect.com/science/article/pii/S0952197600000117
[12] Yaser, E.Hawas. A fuzzy-based system for incident detection in urban
street networks. Transportation Research Part C, 2007, 15:69-95.

Authorized licensed use limited to: Chang'an University. Downloaded on October

307 23,2023 at 09:22:50 UTC from IEEE Xplore. Restrictions apply.