American Journal of Innovative Research and Applied Sciences. ISSN 24295396 I www.americanjiras.com
ORIGINAL ARTICLE
 Marwane Benhadou 1* Amina El gonnouni 2  and  Abdelouahid Lyhyaoui 3 
1. Abdelmalek Essaadi University  Laboratory of innovative technology  Tangier  Morocco 
2. Abdelmalek Essaadi University  Laboratory of innovative technology  Tangier  Morocco 
3. Abdelmalek Essaadi University  Laboratory of innovative technology  Tangier  Morocco 
 Received 26 August 2019   Accepted 28 September 2019   Published 04 October 2019   ID Article  MarwaneRef.1ajira2260919
ABSTRACT
Background: Urban mobility plays an important role in the citizen’s life quality. Good mobility management has a very important impact in facilitating travel and improving the city environment quality (climate). When we talk about mobility we will refer to any mode of motorized displacement or not. In the literature there are many essays that focus on the study of both pedestrian and vehicular mobility. One of the most conflictive points in the city are the roundabouts, and where more time is wasted. Objectives: In this article we will focus on the study of private vehicles mobility within the city of Tangier, and exactly at a specified roundabout. The intersections are a very conflicting point where all the vehicle flows converge. There are two types of intersections, those controlled by traffic lights and those not controlled, the latter are characterized by respecting some rules of priority, and the most important rule is to give way for vehicles that circulate within the roundabout. The most important parameter is the roundabout capacity, which is the hourly flow that enters in each leg (Q_f), that flow is expressed by the annular flow Q_c, and other parameters such as the geometry of the roundabout. The objective will be to estimate the roundabout capacity using SVM algorithm. Methods: We locate the study area through google maps, to identify the roundabout, we model the intersection as a graph form to facilitate the task of data collection, and using a video recording in peak hours we can identify the flow in each leg (Q_f) and the flow inside the roundabout (Q_c). Finally, we use SVM algorithm to estimate the roundabout capacity. Results: The result will be the estimation of traffic flow using SVM algorithm, and capacity simulation at various intervals of the day and in different months of the year. Conclusions: Machine learning algorithm it’s a good tool to model vehicles flow, and the simulations show a good estimate using SVM algorithm. As an improvement for a future study, it will be interesting to compare the result of SVM model with other models, both linear and nonlinear.
Keywords: Roundabout capacity, Support Vector Machine Algorithm, Mobility in Tangier.
1. INTRODUCTION
Urban mobility is causing more and more problems, such as atmospheric and noise pollution that have a serious impact on health [1]. Currently most of the studies focus on sustainable mobility, as a key to solve the problem of displacement within the city [2]. From a simplified approach, you can model a city as a graph, the edges constitute the different streets, and the nodes, corresponds to the streets intersections [3]. There are three types of nodes, a simple intersection with ‘right priority’, a signalized intersection and a roundabout with ‘left priority’ (vehicles entering a roundabout must give way to vehicles in the roundabout) [4]. The intersections in a city, is an essential transit point, when the driver wants to change directions, is also a point of convergence of different streets, causing congestion, which is translated into increasing amounts of pollutants, and vehicles accident.
How can an intersection be examined? Specifically, a roundabout with ‘left priority’. The level of service in a roundabout is identified by the roundabout capacity and the time lost (delay) going through [4]. There are several models and methods to calculate the roundabout capacity.
In these last 10 years, the city of Tangier has known an interesting economic and demographic growth, traducing into more displacements [5]. The roundabout selected to present this study, is specified by its location in the center of the city, the location of attractive centers around it, such as the Mohamed 5 Hospital, the Spanish consulate, it is also a point of convergence of many directions and is a transit point towards the city center, as depicted in figure 2.
2. MATERIALS AND METHODS 2. MATERIALS AND METHODS 2.1 Related works: Many authors have studied the problem of estimating roundabouts capacity, as well as the different parameters that affect security, such as the critical gap, followup time, and the roundabout geometry [6]. There are 2 large families of methods, the empirical methods based on linear regression and the analytical or probabilistic methods based on gap acceptance theory [6]. There are some authors who support their studies using microscopic simulation [6]. RuiJun et al., (2009) have analyzed the difference between singlelane and doublelane roundabouts in China using analytics methods based on M3 distribution [7]. Also Bie et al., (2010) investigates a lanebased method, evaluating the entry capacity for each individual lane using the HCM formula and Kimber model [8]. The research work set up by Wang and Yang (2012) estimate the roundabout capacity by modelling weaving gap acceptance at the weaving section, where they have studied 21 roundabouts in Beijing [9]. In Slovakia the capacity evaluation of roundabout is studied according to Swiss empirical method and German guidelines, evaluating each roundabout by using LOS (Level Of Service) [10]. In another research work, the author provided the different traffic operation at roundabouts such as: distribution of circulation velocity, entry velocity, circulating gap and entry gap [11]. Also, in Argentina, Castellano (2013), uses the roundabout capacity calibration of HCM 2010 model, applying to the local conditions of Cordoba in Argentina in 4 roundabouts, the calibration parameters are the critical gap and the followup interval, to this purpose use ML (maximum likelihood), linear regression and direct measurement [12]. Kang and Nakamura, (2014) analyzed the estimating roundabout capacity taking into account, pedestrian approaching side and queueing vehicles blocked by pedestrians [13]. Ersoy and Celikoglu (2014) estimate roundabout capacity with the incorporate of HCM 2010 method and TRL method [14]. Lochrane et al (2014), provides design recommendation and simulation approach for miniroundabout using FHWA (the Federal Highway Administration), and the only capacity model that is available for analyzing miniroundabouts, is an empiricalbased model from a study by the UK ARCADY (TRL) [15]. Zhaowei et al. (2014) highlights a review of roundabout capacity, estimating the different approach of roundabout capacity such as, empirical regression model, gap acceptance model, weaving theory (Wardrop) and model based on simulation [16]. Agyerman et al. (2015) studied 3 roundabouts in Sunyani in Ghana, using data collection during peak hours in interval of 15 min and using simulation to estimate capacities and LOS. The roundabout capacity is determined by Kimber formula [17]. In another article Arrgu et al. (2015), determined the roundabout capacity using various capacity formulas, such as, gap acceptance model (HCM 2010 USA), German model 2001, empirical regression model (TRRL, UK) and weaving models, also a microscopic simulation using VISSIM [18]. Mahech et al. (2014) examines the entry capacity of a roundabout under different circulating flows, using the HCM 2010 [19]. Ahmed and Rastogi (2016) presents a regression model which is developed to estimate entry capacity, the author has studied different approach of regression based linear and nonlinear models [20]. Also the authors use correlation analysis to show how strongly is a statistical technique, and the results show that the central island diameter is correlated with the entry capacity and develop 2 models, linear and nonlinear models. Another study, Mathew et al. (2017) investigates the suitability of HCM equations for determining the entry capacity of fourlegged roundabout and propose model for entry capacity [21]. Brilon (2016), the critical gap is determined by using MLM (maximum likelihood method) [22].In regards to traffic flow estimation, many authors developed the estimation model associated with flow prediction as clustering methods introduced by Caceres et al. (2012) [23], Kalman filter studied by Chu et al. (2005) [24], artificial neural networks (ANN) developed by Wei and Lee (2003) [25], and prediction mechanism based on the traffic flow spacetime characteristics in Gaoerji (China) road, a oneway lane, using SVM algorithm applied by Mingheng et al. (2013) [26].2.2 Capacity estimation:There are 2 types of intersections, roundabouts, where circulating vehicles had priority over entering vehicles, and the rest of the intersections. The roundabouts with offside priority, are intersections of 2 or more routes, the essential characteristics represented in figure 1. Currently, the roundabouts began to be considered as a sum of 'T' intersections, and the concept of global capacity is replaced by the capacity of each access. Therefore, the capacity is: The greatest number of vehicles that can enter to the roundabout from an access, during a period of time under certain traffic conditions and geometry. According to the different studies, the capacity represents the relationship between the circulating traffic (Q_c) and the maximum that could be incorporated in the entrance (Q_e) [6, 27]. There are two methodologies, empirical methods and analytical (probabilistic) methods. Most countries have adopted their capacity model based on one of these 2 methods, the regression base (United Kingdom) and the gap acceptance theory (Australia) [6, 27]. Figure 1: The figure presents the roundabout geometry and the different influential parameters.2.3 Study area and data collection:The city center area where most of the vehicle flows converge, as shown in the figure 2 (a). The fourleg roundabout is identified by red circle within the study area, is the intersection between Habib Bourguiba avenue, Sidi Amar avenue and Elouchak avenue. We model the roundabout to facilitate data collection as depicted in figure 2 (b). The data collection was done using video recordings, in different months of the year (April, July and October) and during a week of every month from Monday to Sunday during 7 time intervals in 30 min of recording {7h458h15, 9h4510h15, 11h4512h15, 13h4514h15, 15h4516h15, 17h4518h15, 19h4520h15}.Figure 2: The figure presents the roundabout site (a) and modeling (b).2.4 Support vector machine model:Machine learning model learns from the past input data and makes future prediction as output. We can divide machine learning algorithms into three main groups based on their purpose: a supervised learning, unsupervised learning and reinforcement learning [28]. Support Vector Machine algorithm abbreviated as SVM is specific to supervised learning is a powerful method for regression and classification introduced by Vapink (1995) [29]. The SVM maps the input data (
R
2
{R} ^ {2}
) into
(
R
3
)
( {R} ^ {3} )
space via a nonlinear map, constructing a hyperplane with maximum margin. The following shows the SVM algorithm [30]:Consider a given training set
{
x
i
,
y
i
:
i
=
1,
⋯
l
}
size 12{ left lbrace x rSub { size 8{i} } ,y rSub { size 8{i} } :i=1, dotsaxis l right rbrace } {}
randomly and independently generated from an unknown function, where
x
i
∈
X
⊆
R
n
,
y
∈
Y
⊆
R
size 12{x rSub { size 8{i} } in X subseteq R rSup { size 8{n} } , y in Y subseteq R} {}
and
l
size 12{l} {}
is the total number of training data.The SVM approximates the unknown function:
f
(
x
)
=
⟨
w
,
Φ
(
x
)
⟩
+
b
size 12{f \( x \) = langle w,Φ \( x \) rangle +b} {}
(1)Where
⟨
,
⟩
size 12{ langle , rangle } {}
is the dot product,
w
size 12{w} {}
and
b
size 12{b} {}
are the estimated parameters and
Φ
size 12{Φ} {}
is a nonlinear function is used to map the original input space
R
n
{R} ^ {n}
to high dimensional feature space. The optimization goal of standard SVM is formulated as:Minimize
1
2
‖
w
‖
2
+
C
∑
i
=
1
l
(
ξ
i
+
ξ
i
)
size 12{ { {1} over {2} } ldline w rdline rSup { size 8{2} } +C Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( ξ rSub { size 8{i} } +ξ rSub { size 8{i} } rSup { size 8{*} } \) } } {}
(2)Subject to:
y
i
−
⟨
w
,
φ
(
x
i
)
⟩
−
b
≤
ε
+
ξ
i
,
size 12{y rSub { size 8{i} }  langle w,φ \( x rSub { size 8{i} } \) rangle  b <= ε+ξ rSub { size 8{i} } ,} {}
⟨
w
,
φ
(
x
i
)
⟩
+
b
−
y
i
≤
ε
+
ξ
i
,
size 12{ langle w,φ \( x rSub { size 8{i} } \) rangle +b  y rSub { size 8{i} } <= ε+ξ rSub { size 8{i} } rSup { size 8{*} } ,} {}
ξ
i
,
ξ
i
≥
0,
i
=
1,
…
…
,
l
.
size 12{ξ rSub { size 8{i} } rSup { size 8{*} } , ξ rSub { size 8{i} } >= 0, i=1, dotslow dotslow ,l "." } {}
Where the constant
C
>
0
C>0
determines the tradeoff between the flatness of
f
f
and the amount up to which deviations larger than are
ε
ε
tolerated and
ξ
i
,
ξ
i
size 12{ξ rSub { size 8{i} } rSup { size 8{*} } , ξ rSub { size 8{i} } } {}
are slack variables and they are introduced to accommodate, respectively, the positive and the negative errors on the training data.
ε
ε
insensitive cost function:
0
if

ξ

<
ε

ξ


ε
otherwise

ξ

ε
:
=
{
size 12{ lline ξ rline rSub { size 8{ε} } :=alignl { stack {
left lbrace 0" if " lline ξ rline <ε {} #
right none left lbrace lline ξ rline " "ε" otherwise" {} #
right no } } lbrace } {}
Maximizing the margin m is a quadratic programming problem QP and can be solved by its dual problem by introducing Lagrange multipliers
α
i
,
α
i
,
β
i
,
β
i
size 12{α rSub { size 8{i} } , α rSub { size 8{i} } rSup { size 8{*} } , β rSub { size 8{i} } , β rSub { size 8{i} } rSup { size 8{*} } } {}
:
L
P
(
w,
ξ
,
ξ
,
α
,
α
,
β
,
β
)
=
1
2
⟨
w
.
w
⟩
+
C
.
∑
i
=
1
l
(
ξ
i
+
ξ
i
)
−
∑
i
=
1
l
α
i
.
⟨
w
,
φ
(
x
i
)
⟩
−
y
i
+
b
+
ε
+
ξ
i
)
size 12{L rSub { size 8{P} } \( "w,"ξ,ξ rSup { size 8{*} } ,α,α rSup { size 8{*} } ,β,β rSup { size 8{*} } \) = { {1} over {2} } langle w "." w rangle + C "." Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( ξ rSub { size 8{i} } +ξ rSub { size 8{i} } rSup { size 8{*} } \) }  Sum cSub { size 8{i=1} } cSup { size 8{l} } {α rSub { size 8{i} } "." langle w,φ \( x rSub { size 8{i} } \) rangle  y rSub { size 8{i} } +b+ε+ξ rSub { size 8{i} } \) } } {}
−
∑
i
=
1
l
α
i
.
(
y
i
−
⟨
w
,
φ
(
x
i
)
⟩
)
−
b
+
ε
+
ξ
i
)
−
∑
i
=
1
l
(
β
i
.
ξ
i
+
β
i
.
ξ
i
)
size 12{  Sum cSub { size 8{i=1} } cSup { size 8{l} } {α rSub { size 8{i} } rSup { size 8{*} } "." \( y rSub { size 8{i} }  langle w,φ \( x rSub { size 8{i} } \) rangle \)  b+ε+ξ rSub { size 8{i} } rSup { size 8{*} } \)  Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( β rSub { size 8{i} } "." ξ rSub { size 8{i} } +β rSub { size 8{i} } rSup { size 8{*} } "." ξ rSub { size 8{i} } rSup { size 8{*} } \) } } } {}
(3)By minimization the Lagrangian with respect to the primal variables we obtain:
w
=
∑
i
=
1
l
(
α
i
−
α
i
)
.
φ
(
x
i
)
size 12{w= Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) } "." φ \( x rSub { size 8{i} } \) } {}
(4)And
∑
i
=
1
l
(
α
i
−
α
i
)
=
0
,
size 12{ Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) } =0," "} {}
0
≤
α
i
≤
C
,
size 12{0 <= α rSub { size 8{i} } <= C,} {}
0
≤
α
i
≤
C
,
size 12{0 <= α rSub { size 8{i} } rSup { size 8{*} } <= C,} {}
i
=
1,
…
…
,
l
size 12{i=1, dotslow dotslow ,l} {}
The dual problem is obtained by introducing (3.1) in (3) and it is expressed as:
maximize 
1
2
∑
i
,
j
=
1
l
(
α
i
−
α
i
)
.
(
α
j
−
α
j
)
.
K
(
x
,
x
i
)
+
∑
i
=
1
l
(
α
i
−
α
i
)
.
y
i
−
∑
i
=
1
l
(
α
i
−
α
i
)
.
ε
size 12{"maximize " { {1} over {2} } Sum cSub { size 8{i,j=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) "." \( α rSub { size 8{j} }  α rSub { size 8{j} } rSup { size 8{*} } \) } "." K \( x,x rSub { size 8{i} } \) + Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) "." y rSub { size 8{i} } }  Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) "." ε} } {}
(5)Subject to :
∑
i
=
1
l
(
α
i
−
α
i
)
=
0
,
size 12{ Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) } =0,} {}
0
≤
α
i
≤
C
,
size 12{0 <= α rSub { size 8{i} } <= C,} {}
0
≤
α
i
≤
C
,
size 12{0 <= α rSub { size 8{i} } rSup { size 8{*} } <= C,} {}
i
=
1,
…
…
,
l
size 12{i=1, dotslow dotslow ,l} {}
Finally, the nonlinear function is obtained as:
f
(
x
)
=
∑
i
=
1
l
(
α
i
−
α
i
)
.
K
(
x
,
x
i
)
+
b
size 12{f \( x \) = Sum cSub { size 8{i=1} } cSup { size 8{l} } { \( α rSub { size 8{i} }  α rSub { size 8{i} } rSup { size 8{*} } \) "." } K \( x,x rSub { size 8{i} } \) +b} {}
(6)Where
K
(
x
i
,
x
j
)
=
⟨
φ
(
x
i
)
,
φ
(
x
j
)
⟩
size 12{K \( x rSub { size 8{i} } ,x rSub { size 8{j} } \) = langle φ \( x rSub { size 8{i} } \) ,φ \( x rSub { size 8{j} } \) rangle } {}
is defined as the kernel function. Any function that satisfies Mercer’s theorem can be used as the kernel function.
3. SVM SIMULATION FOR ROUNDABOUT CAPACITY PREDICTION3. SVM SIMULATION FOR ROUNDABOUT CAPACITY PREDICTION3.1 The input matrix and the output vector to apply the SVM method:We construct the input matrix and the output vector to apply the SVM method:The input matrix:For the input, we used a matrix that contains the following elements: [Q_c, day, month, hour, section].For the day we used the numbers: 1,2, 3,..7 to identify Monday, Tuesday,…Sunday.For months we used 4, 7, 10 to identify, April, july, October.For hours we used numbers 1,2,3,4… to identify time interval 7h458h15, 9h4510h45…7h458h15.We use 1,2,3,4 to identify fourleg roundabout A,B,C,D.The size of this matrix is [11200 lines, 5 columns].The output vectorFor the output vector: contains capacity prediction Qf, its size is: [11200 lines, 1 column]Construction of test elements and trainingThe input matrix and the output vector are divided into two parts: one for the training and the other for the test, whose test represents 30% of the input matrix, chosen in a random manner (and the same thing for the output).The kernel used is RBF, three parameters associated should be determined (𝐶, 𝜀, 𝛾). with C=100, ε=0,1 and for RBF the parameter is 0,2.3.2 SimulationsFor simulations, we represent each month in a separate table and for each hourly interval during the day, as seen in the figures 3,4 and 5. The estimated data is represented with red color and the observed data with green color, using Matlab simulation. The aim is to approximate the prediction performance based on the knowledge of the training set.The results show that the methodology makes good use of historical data, with the Mean Absolute Relative Error MARE=13,61%.Month: April
Time interval 7h458h15Time interval 9h4510h15

Time interval 11h4512h15Time interval 1h452h15

Time interval 3h454h15Time interval 5h456h15

Time interval 7h45 8h15
Figure 3: The figure presents the the simulation result for April in 7 time intervals.In the simulation result for July, the results show that the methodology makes good use of historical data, with the Mean Absolute Relative Error MARE=12,2%.Month: July
Time interval 7h458h15Time interval 9h4510h15

Time interval 11h4512h15Time interval 1h452h15

Time interval 3h454h15Time interval 5h456h15

Time interval 7h45 8h15
Figure 4: The figure presents the simulation result for July in 7 time intervals.In the simulation result for October. The results show that the methodology makes good use of historical data, with the Mean Absolute Relative Error MARE=14,43%.Month: October
Time interval 7h458h15Time interval 9h4510h15

Time interval 11h4512h15Time interval 1h452h15

Time interval 3h454h15Time interval 5h456h15

Time interval 7h45 8h15
Figure 5: The figure presents the simulation result for October in 7 time intervals.Now we can represent the total capacity estimation as shown in figure 6. Figure 6: The figure presents the total roundabout capacity estimation. From the results, it can be seen that the prediction curves agree well with the actual observing data.4. CONCLUSION Urban mobility plays a very important role in the lives of citizens. Good mobility management influences time travels. The studied area is characterized by its central location, such that the vehicles flow come from all directions. The methodology is based first on data collection using video recording in 3 months of the year (April, July and October) throughout the week. In order to predict traffic flow or roundabout capacity, this paper proposed a SVM model for the capacity prediction. The results achieved show a good representation of our data through SVM and a good ability for capacity prediction. In further study we can compare the SVM model with other linear and nonlinear models, also we can simulate the different methods of capacity calculation, such as the French method, the English model (Kimber), HCM 2010, the German ... to see who fits more to our data.5. REFERENCES 1. Compilador A. B. Congestión de tránsito el problema y cómo enfrentarlo. Santiago de Chile, 2003.2. Wefering F., Rupprecht S., Bührmann S,, BöhlerBaedeker S. Developing and implementing a sustainable urban mobility plan, guidelines. European union 2013.3. Sheffi Y. Urban Transportation Networks. Massachusetts institute of technology, PrenticeHal, 1985.4. Roundabouts: An information guide. US Departement of Transportation, Federal Highway Administration, Publication No. FHWARD00067, 2000.5. Statistical Yearbook. 2000 to 2013, the TangierTétouan region.6. Roger P. R, Elena S. Prassas, William R. M. Traffic engineering. fourth edition.7. RuiJun G., BoLiang L., WanXiang W. The Iterative Calculation of Full Capacity of Roundabouts. Second International Symposium on Information Science and Engineering, DOI 10.1109/ISISE.2009.358. Bie J., Lo H.K., Wong S.C. Capacity evaluation of multilane traffic roundabout. J. Adv. Transp. 2010; 44: 245–255. DOI: 10.1002/atr.124.9. Wang W., Yang X. Research on Capacity of Roundabouts in Beijing. 8th International Conference on Traffic and Transportation Studies, Changsha, China, August 1–3, 2012.10. Gavulova A., Drliciak M. CAPACITY EVALUATION OF ROUNDABOUTS IN SLOVAKIA. Transport and Telecommunication. 2012; 13(1): 1–1011. Liu Y., Guo X., Kong D., Liang H. Analysis of Traffic Operation Performances at Roundabouts. 13th COTA International Conference of Transportation Professionals (CICTP 2013.)12. Castellano G.J.F. CALIBRACIÓN DEL MODELO DE CAPACIDAD DE ROTONDAS DEL HCM2010 A CONDICIONES LOCALES: CASO CÓRDOBA, ARGENTINA. Congreso de Pesquisa en Ensino em Transportes, 2013.13. Kang N., Nakamura H. An Estimation Method of Roundabout Entry Capacity Considering pedestrian impact. The 9th International Conference on Traffic & Transportation Studies (ICTTS’2014).14. Ersoy M., Çelikoglu H.B. CAPACITY ANALYSIS ON MULTILANE ROUNDABOUTS: AN EVALUATION WITH HIGHWAY CAPACITY MANUAL 2010 CAPACITY MODEL. Pamukkale Univ Muh Bilim Derg. 2014; 20(6), 225231,.15. Lochrane T.W.P, Kronprasent N., Bared J., Dailey D.J., Zhang W. Determination of MiniRoundabout Capacity in the United States. American Society of Civil Engineers. 2014. DOI: 10.1061/(ASCE)TE.19435436.0000700.16. Zhaowei Q., Yuzhou D., Xianmin S., Yan X. Review and Outlook of Roundabout Capacity. J Transpn Sys Eng & IT. 2014; 14(5): 15−22.17. Agyerman S., Abeka H., Asiedu S.B. Capacity and Performance Analysis of 3 Roundabouts in Sunyani. International Journal of Science and Research. 2015; 4(10).18. Arrgu R., Gaddam H.K., Vanum L.D., Rao K.R. Comparative evaluation of roundabout capacities under heterogeneous traffic conditions. J. Mod. Transport. 2015; 23(4):310–324. DOI 10.1007/s4053401500898.19. Mahech S., Ahmad A., Rastogi R. An approach for the estimation of entry flows on roundabouts.11th Transportation Planning and Implementation Methodologies for Developing Countries TPMDC, 2014, 1012 December 2014, Mumbai, India.20. Ahmad A., Rastogi R. Regression model for entry capacity of a roundabout under mixed traffic condition – an Indian case study. The International Journal of Transportation Research. 2016. DOI: 10.1080/19427867.2016.1203603.21. Mathew S., Dhamaniya A., Arkatkar S.S., Joshi G. Roundabout Capacity in Heterogeneous Traffic Condition: Modification of HCM Equation and Calibration. 20th EURO Working Group on Transportation Meeting, EWGT 2017, 46 September 2017, Budapest, Hungary, Transportation Research Procedia . 2017; (27): 985–99222. Brilon. Some remarks regarding the estimation of critical gaps, transportation research record . Journal of the transportation research Board, no 2553, Transportation research board of the national academics, Washington D.C, 2016.23. Caceres N., Romero L.M., Benitez F.G. Estimating traffic flow profiles according to a relative attractiveness factor. 15th meeting of the EURO Working Group on Transportation, EWGT 2012. Procedia  Social and Behavioral Sciences. 2012 ; (54) 1115 – 1124.24. Chu, L., Oh, JS., & Recker, W. (2005). Adaptive Kalman Filter Based Freeway Travel Time Estimation. Transportation Research Board 84th, Annual Meeting, CDROM, paper no. 1118.25. Wei, C.H., & Lee, Y. (2003). Development of freeway travel time forecasting models using artificial neural networks. 10th World Congress on Intelligent Transportation Systems, Madrid, CDROM Paper.26. Z. Mingheng, Z. Yaobao, H. Ganglong, C. Gang. Accurate Multisteps Traffic Flow Prediction Based on SVM. Mathematical Problems in Engineering; 2013. 19. http://dx.doi.org/10.1155/2013/418303.27. HCM 2010. hign capacity manual .28. Cortes C., Vapnik V.Ñ. Support vector networks. Machine Learning. 1995; (20): 273297, 29. Vapnik V.Ñ. The nature of statistical learning theory. New York: SpringerVerlag, 1995.30. Cristianini N. , ShaweTaylor J. Support vector machines and other kernelbased learning methods. Cambridge University Press, Cambridbe MA, ISBN 0521780195, 2000.