CONCEPTUAL COST ESTIMATION OF ROAD PROJECTS IN ETHIOPIA

USING NEURAL NETWORKS

Nardos Tadesse and Abebe Dinku
School of Civil and Environmental Engineering
Addis Ababa Institute of Technology, AAU

ABSTRACT
The higher amount of budget allocated to the
The rapid technological changes and advances road construction industry stresses the
in all business sectors strongly impose requirement of applying advanced planning
construction managers to facilitate their work tools for project completion within estimated
through advanced software applications cost.
available to simplify different tasks. A
research has shown that inaccurate cost The Government of Ethiopia formulated the
estimation method is among the cause of cost Road Sector Development Program (RSDP I to
overrun in the Ethiopian construction industry. IV) in 1997 and disbursed 48.11 Billion Birr in
This article presents conceptual cost the years 2011 and 2012 for the RSDP IV [2].
estimation model developed using neural A research conducted by ERA Research Center
networks for federal road projects of Ethiopia. on causes of cost overrun on Ethiopian Federal
The conceptual cost estimation model Road projects in 2016 showed that 10 to 21%
developed has a mean absolute percentage of cost overrun is recorded [3]. This shows the
error of 32.58% trained with only 48 requirement of measurement to be taken to
exemplars. If the model is developed with the improve cost estimate reliability.
provision of enough data set to represent the
road project with all-state of affairs, it is A research revealed that inaccurate cost
forecasted to improve the estimating capability estimation method and poor planning are
of financers, employers and consultants. In among the factors contributing to cost overrun
addition, a friendly user interface is built to in the Ethiopian construction industry [4].
enable the utilization of the model developed
easily and this article presents this interface
with an example of actual road project in In addition, the review of best practice, which is
Ethiopia. The findings of this study indicate the research conducted by U.S. department of
the prospect of application of neural network transportation Federal highway administration,
for cost estimation during early phase of the showed that cost estimation programs and
project development for Ethiopian road quality control & assurance programs are
projects. developed in light of addressing cost overrun
issues [1].
Key Words: Conceptual Cost Estimation,
Neural Network, Parameters, Reliable estimate This paper aims to use a similar approach and
develop a conceptual cost estimation model for
INTRODUCTION Ethiopian road projects through the application
Reliable cost estimates are necessary for of Neural Network. Therefore, identification of
responsible cost management at every stage of factors that affect total cost, which are to be
a project. The application of advanced used as parameters in training a Neural
conceptual cost estimation tools in Ethiopian Network model, is the first objective.
Construction industry is at an infant stage and
requires localized researches. Cost overrun of The chief goal is to support decision makers
infrastructure projects has become a common through increased reliability of estimated cost
problem for the construction industry as of road construction projects at early stages of
recognized by industry researchers and project development by providing cost
practitioners [1]. estimation tool.

17
Journal of EEA, Vol. 35, October, 2017
LITERATURE REVIEW

Conceptual Cost Estimation

A conceptual estimate is made during early
phase of project development without detailed
design and engineering data, and with limited
information on project scope. Therefore,
considerable experience and judgment are
required to obtain a dependable approximate
estimate for the cost.

Estimating is the heart of the cost engineer‟s

work and consequently, it has received
appropriate attention over the years. The first
function of a conceptual estimate is to tell the
owner about the anticipated cost, thus
presenting useful information for the owner in
contemplating the project feasibility and They should be limited to those items that
further development [6]. Conceptual and comprise the largest cost percentage of the
preliminary estimates are made for several project, typically that 20% of items of work
reasons, including [7] [8]: that account for 80% of project cost. The cost
 Feasibility studies of the remainder of estimate line items can be
 To establish a basis for financing, determined using Historical Bid-Based
appropriation of funds or validation of Estimate methods [11].
project budgets
 Appropriation of the project scope and Risk-based
selection from alternate design and Risk-based estimate methods include complex
investment probabilistic analysis with identification of
 Presentation of bids and tenders which is opportunities and threats related to cost,
to establish a benchmark for a construction schedule, and events related to the project. It
bid uses different techniques, such as historical
data, cost based estimating, and the best
Cost Estimating Methods judgment of subject matter experts for given
Cost Estimating methodologies fall into one of the types of work [5].
following four categories discussed below [5].
Information Available during Conceptual
Historical bid-based Estimation Project Development Level

Historical bid-based methods are commonly The information available is one of the
used to develop engineer‟s estimates, and are distinctive features of the different types of
applied when the project design progressed cost estimates. The estimate level among the
enough to provide quantities of unit of works. different types of estimates available does not
Historical unit costs are applied to estimate the give clear and definite boundary. Due to this
total cost of each work items [5]. Cost based reason, there are some literatures that use the
Estimation terms conceptual and preliminary estimate
Cost-based estimate methods are based on interchangeably. During the conceptual phase,
calculation of unit costs for each project with general information about the project will be
details of labor, equipment, material and available such as the road length, road width
specialty contractor estimation for each item from standards, the location of the project and
of work. Historical bid data are not utilized the project scope.
[5]. Cost-based estimates require significant
effort, time, and estimator experience to
prepare.

18
Journal of EEA, Vol. 35, October, 2017
And a parametric cost estimation technique Adaptive: Neural networks have a built-in
can be deployed from historical costs of capability to adapt their synaptic weights to
similar past projects. During the preliminary changes in the surrounding environment.
phase, preliminary quantities of earthwork and i. Contextual information: contextual
pavements are available by the use of digital information is dealt with naturally by a neural
terrain model and different maps. Figure 1 network.
below shows the available information, ii. Fault tolerance: a neural network has the
difficulty level and expected percentage error potential to be inherently fault tolerant.
for different types of cost estimates.  To predict acceptability of a new
technology, cash flow, construction
The identification of the influential factors or demand, construction budget performance
parameters affecting total project construction and construction cost
cost is necessary to develop cost estimating  To investigated the overhead cost
model. The input parameters identified practices of construction companies in
through literature review of road construction Saudi Arabia
cost estimation are summarized below in table  To develop cost estimating model for the
1. structural systems of reinforced concrete
buildings in the sense that its performance
Artificial Neural Network is degraded gracefully under adverse
operating. Although the artificial neural
Artificial intelligence is defined as: "The study networks have many advantages, on the other
of how to make computers do things at which, hand their accuracy highly depends on the
at the moment, people are better" Rich and quality of the trained data and the ability of the
Knight, 1991 developer to choose truly representative
sample inputs. In addition, trial and error
method used to obtain the formula to decide
Artificial Intelligence is the branch of
what architecture of ANN should be used to
computer science that is concerned with the solve the given problem and which training
automation of intelligent behavior and seeks to algorithm to use.
explain and emulate intelligent behavior in
terms of computational processes [15]. Application of ANNs in Construction
Artificial Neural Networks (ANNs) are a Engineering
functional abstraction of the biologic neural
structures of the central nervous system The most common application of ANNs in
[16].ANN generally learns from “experience”, the construction management area is
rather than being explicitly “programmed” prediction. Few of the applications of the
with rules like in conventional artificial ANN in construction management for
intelligence [17]. prediction and estimation in construction
engineering are given below [9, 16]:
Benefits and limitations of Artificial  Model for predicting increase in time
Neural Network and cost of construction projects in
The use of neural networks offers the Egypt
following useful properties and capabilities  To forecast actual cost of a project
[18]: based on the earned value
management system
Input-output mapping: a popular paradigm of
 To estimate conceptual cost of Libyan
learning called supervised learning involves
Highway Projects
the modification of the synaptic weights (free
parameters) of a neural network by applying a
The literatures show the vast application
set of training samples. Thus the network
of ANN in the construction industry
learns from the samples by constructing an
worldwide and the need of the Ethiopian
input-output mapping for the problem at
construction industry to advance in order
hand.
to cope up with the rapid and dynamic
advancement of technology is undeniable.

19
Journal of EEA, Vol. 35, October, 2017
Table 1: Previously used parameters by other research papers [9, 12, 13, 14]

Year
Research description Input parameters
1998 Neural network approach was used to Project type, project scope, year, construction
determine highway construction cost season, location, duration, size, capacity, water
in Canada body, and soil condition.
2005 Multiple regression model and ANN Predominant Work Activity (Asphalt or
model were used to estimate the cost Concrete), Work Duration, Pavement Width,
at conceptual phase of highway Shoulder Width, Ground Rise Fall, Average Site
projects in Poland and Thailand. Clear/Grub, Earthwork Volume, Surface Class
(Asphalt or Concrete), and Base Material
(Crushed Stone or Cement Stab).
2013 Parametric Cost Estimation of Road Project scope, Water networks, Pavement type
Projects Using Artificial Neural (asphalt or interlock), Pavement area, Length of
Networks in Palestine the road, Length of the curbstones Lighting
networks, Sewage networks
2013 Conceptual Cost Estimate of Road Earthwork; cut and fill (m3), Base works (m2),
Construction Projects in Saudi Asphalt works (m2), Road length, Road width
Arabia by developing regression (m)
models to predict the total
construction cost of a road project in
the early phases.
2014 Testing regression models to estimate Earthwork; cut, fill, and topping quantities (m3),
costs of road construction projects, to Sub base and Base coarse quantity (m3), Asphalt
develop early cost estimating models quantity (m2), Road width (m), Road length (m)
of road construction projects in
Ethiopia

20
Journal of EEA, Vol. 35, October, 2017
RESEARCH METHODOLOGY parametric estimation. The collected data were
compiled from Ethiopian Road Authority
The research methods applied for this research Management System (ERAMS), contract
are correlational descriptive and quantitative documents and reports. A total of 58 data were
research methods. Correlational studies trace collected for the model development. The data
relationships among two or more variables in were then analyzed and put in table format for
order to gain greater situational insight. simplicity and avoidance of double use of
Therefore to identify the parameters to be used specific project.
for developing the cost estimation model
correlational descriptive research method was The succeeding step was development of the
adopted. The quantitative research method was neural network model by training the program.
adopted to develop the model that estimates Then the model was tested and sensitivity
cost. analysis was carried out. A friendly interface
The study area of this research is Ethiopia was then created to link the neural network
excluding the capital city, Addis Ababa. Addis model with users. Finally, conclusions were
Ababa is the most urbanized, highly congested made and recommendations were forwarded in
and populated area fading the resemblance and subgroup of the party aimed to be addressed.
similarity of the road construction cost among
the other areas. Most of the road networks are MODEL DEVELOPMENT
built up which has a significant influence of
right of way issues affecting construction The cost estimating models are developed in
periods and work progress which in turn series of steps as shown in Figure 2.
influences the project cost. The road projects
taken as historical feed to develop the The preprocess of data in this paper is
parametric cost estimation are from the North, designated to arranging the data in rows and
South, East, West and Central region. The columns, encoding the data, randomizing the
central region includes road projects in data and conveying the total project costs to a
250KM radius from Addis Ababa city. The common base year. NeuroSolutions creates a
principal tools applied are Microsoft Excel new breadboard with typical elements used for
2013, Neuro Solutions version 7.1.0.0, and neural network configurations chosen and the
Visual Basic Application. breadboard is saved. The breadboard created is
shown in Figure 3. The chosen models for this
Research procedures research include linear regression, multilayer
perceptron and generalized feed forward. NNs
The study started with an intensive literature are then trained using available data to
review and consequently data were collected. understand the underlying pattern. During
The data collection was conducted in two training, both the inputs, representing problem
phases as preliminary data collection and final parameters and outputs, representing the
data collection. The preliminary data was solutions are presented to the network
collected through 13 experts‟ interview in normally for thousands of cycles. At the end of
open and close ended questions. The experts each cycle, or iteration, the network evaluates
interviewed include Highway Engineers, the error between the desired output and actual
Contract Administrators, Quantity Surveyors, output.
University Lecturers and from client side,
Project Team Leaders

and Counterparts from ERA. The analysis of

the preliminary data collected was carried out
for the sole purpose of identifying and
determining input parameters. This stage was
imperative as not to escape any important
parameter that may affect cost of road
projects. The next step was the collection of
final data to be used as historical basis for

22
Journal of EEA, Vol. 35, October, 2017
 management System) software did not include
Then this error is used to modify the this parameter as an output report of the
connection weights according to the training system even though the data was initially feed
algorithms used [9]. Back propagation is the to develop the software.
type of algorithm adapted, which is a common
method of teaching artificial neural networks The two project delivery systems
and uses supervised learning. predominantly used in the Ethiopian road
The trained network is tested with the data set construction are design-bid-build (DBB) and
designated for testing and the network with the design-build (DB). The estimation structures
best performance is chosen. The performance of the two are totally different and therefore
measures include Mean Square Error (MSE) project delivery system was not used as a
and Correlation Coefficient (r). Finally a parameter. Therefore, the scope of this study is
sensitivity analysis was carried out. DBB projects for the reason that only DBB
project‟s data were collected.

ANALYSIS AND DISCUSSION Project location required further study on how

to classify the location of the study area. The
The analysis and discussion is comprised of study area of this paper is Ethiopia excluding
three subsections over which preliminary data Addis Ababa. The concerns taken into
collected, final data collected and the results of consideration for classification purpose in this
the model developed are analyzed and study are climate and sources of four
discussed in detail. construction materials, which affect
construction cost and time. The study was
Analysis and Discussion on Preliminary started with an assumption and then the
Data Collection assumption was proven. The assumption was
for the location classification to be the same as
The preliminary data collected was analyzed ERA‟s regional administrative organization as
with the aim of identifying the influential North, South, East, West and Central region.
parameters affecting road project cost during
conceptual stage. One of the findings of the The first point of consideration taken to prove
interview is that two estimate models can be the assumption is climate. Rainfall is the major
predicted during the early phase of the project, climate measure, as discussed in ERA manual
which are conceptual and preliminary 2013. The rainfall distribution of Ethiopia
estimates. This article only focuses on the supports the justification of the assumption as
conceptual estimation level. The detailed shown with the broken lines on the map below
analysis, discussion and results of the of Figure 4.
preliminary cost estimation model are given
on the research paper submitted to Addis
Ababa Institute of Technology [19]. The
parameters affecting conceptual cost of
Ethiopian road projects are briefly discussed.

The parameters identified for the conceptual

estimation model are Project location, Project
scope or type, Wearing Surface Class, Road
length and width, Terrain classification,
Project delivery system and Structure
intensiveness.

Terrain classification and Structural

intensiveness are discarded and taken as a
limitation and are recommended for future
The second accounted point on proving the
studies because they could not be compiled for
assumption is the sources of construction
all project data collected. This is because the
materials. The travel distance and cost for
ERAMS (Ethiopian Road Authority

23
Journal of EEA, Vol. 35, October, 2017
transportation of materials depends on the to period of road construction staring from
distance between the material manufacturing commencement to completion. Construction
(importing site) and the construction sites. For year in this research refers to the year of
instance a road project located in Dire Dawa construction commencement which is
will have less bitumen transportation cost than identified as a parameter on this study based
a project located in Gambella. This can be on previously executed researches. The year of
easily seen on Figure 5.The materials taken construction is related to price escalation
into consideration are asphalt bitumen, taking into account the supply and demand
cement, fuel and steel reinforcement bars. balance of the market and inflation rate. The
year of construction was used to convey the
Cement is one of the materials used for total project costs to a common base year of
concrete production mostly for structures in 2006 through the application of consumer
road construction. According to a report of the price index (CPI) from Central Statistical
presentation on the 7th Africa Cement Trade Agency of Federal Democratic Republic of
Summit in 2015, the Ethiopian cement Ethiopia, based on available data as presented
industry experiences insignificant import that on Table 2. The road length and width were
is only limited to special cements [21]. identified as an input parameter during
Therefore the supply of cement for the road literature review and fully supported by the
construction is from local manufacturers. interview. The road length was used to
Dangote, Derba, National, Messebo, and develop the estimation model. But the width
Mugher cement factories are the major was castoff due to the reason that data
suppliers. Figure 5 shows the distribution of collected on road width shows that more than
the factories is not uniform thought out the 90% of the data is similar. Therefore it was
country which results in different cement discarded and not used for the model
transportation cost along the regions. This development.
similarly supports the assumption of location
categorization as north, south, east, west and
central region. Table 2: Consumer Price index for
construction projects.

Consumer Price
Year Index
2006 100.0
2007 112.0
2008 136.6
2009 160.0
2010 195.7
2011 219.4
2012 245.1
2013 267.9
2014 292.3

From the above discussions, location is

determined to be used as influential Analysis and Discussion on Final Data
parameters for conceptual road project cost Collected
estimation in the subdivision of south, north,
east, west, and central regions. The project The final data collection resulted in production
scopes that are taken into account by this study of set of data to be used as historical feed for
are new construction of roads and upgrading analogous parametric estimation. Totally 58
of existing roads. Construction duration refers data were collected and out of these 48 data

24
Journal of EEA, Vol. 35, October, 2017
were used for conceptual estimation model
development. The rejected data were not able
to be used for model development due to
incompleteness of data, inconsistency of data,
and because of being on-going project data.
The percentage distributions of the data set in
terms of location are shown in Figure 6.

The study took the actual completion date and

commencement date of the projects to
compute the duration, therefore the actual
period of construction is taken. The project
road length is the other parameter
encompassed in the model. The length
distribution of the project data collected is
shown in Figure 8 after being classified into
four.

Depending on the percentage distribution of

data set for location, the north area outshines
accounting for the 35% and on the other hand
the data set from east region accounts for 10%
of the total data set.

Figure 7 shows the percentage of data

collected with respect to surface class, which
are classified as Asphalt Concrete, Double
Bituminous Surface Treatment and Gravel
Wearing Course

Discussion on Model development epoch when the cross validation error is

minimum for conceptual cost estimation
The percentage of row tags have been varied
while training the network to get best accuracy Table 3: Input/output tags and Percentage of
of the models with respect to the performance raw tags
measures given in testing reports. The data %
tags of the two models are shown in Tables 3 Input Output Row of
tags tag tags row
below. tags
Project Location
Training Result discussion Project Scope Training
Total Road 70%
Wearing Surface Cross
Construction 10%
After building a neural network breadboard Class Validation
Cost 20%
the next task performed was training the Road Length Testing
Duration
network to develop a model. The training
result presented below in Figure 9 shows the
best network weight which is saved at the

25
Journal of EEA, Vol. 35, October, 2017
model developed. As it is shown in the figure,
the training of network stopped after training
and cross validation means square errors
(MSE) have reached the minimum and kept
constant for continuing trials. The training
result of the model shows the MSE
approached minimum during the early phase
of trials or epoch. After many trials were
performed, the neural network configuration
or topology with the best performance
measure for the conceptual estimation model
is Regression General Feed forward network
with the architecture of GFFR-1-O-M. The
architecture represents a Regression General Coefficient of determination denoted as r2 of
Network with one hidden layer, Momentum network output verses desired graph, which is
Gradient search and On-line learning updates equal to 0.82, shows that the network put fits
method. The networks trained were then tested the desired output and moves in the same
with the provided testing data sets to the direction, as given on Figure 11. With the
performance of the models. provided number of data set, these are
promising results towards the possibility of
better accuracy with provision of more data
points.

Sensitivity Analysis Result

Test Results discussion
The sensitivity analysis was carried out for the
The model was trained with 48 exemplars and estimation models that was trained and tested.
33 of the exemplars were used for training; 5 To carry out the sensitivity analysis the inputs
of the exemplars were used for cross are varied between its mean ±1 of standard
validation to produce better output for unseen deviations while all other inputs are fixed at
examples; and 10 of the exemplars were used their respective means. This is done for 50
for testing. Figure 10 below shows the desired steps to each side. A report is then generated
output and network output of the conceptual which summarizes the variation of each output
cost estimation model developed converge in with respect to the variation in each input
the same direction in close proximity. parameter. The report is presented in Figure 12
below.

26
Journal of EEA, Vol. 35, October, 2017
 Structuring User Interface

The friendly user interface is created to use

of the model by simply entering the input
parameters and receive the predicted total
cost is shown in Fig. 13.

As it is seen on the chart above construction

period has the lowest effect on the model;
however this parameter was not eliminated or
discarded due to the reason that the model was
developed for projects completed over
different period of years. Project scope and
road length have the highest effect of the
model. With the number of exemplars
provided under consideration, the sensitivity
analysis result shows that all the parameters
used have indisputable impact on the total
with an example of an upgrading road
construction cost.
project cost estimate. The actual contract
Discussion on the Expected Accuracy of
amount for the construction of the road
the model
project is 1,142,567,660.95 ETB. Therefore,
the mean absolute percentage error
The main performance measures this study
calculated using the actual and desired
used to evaluate the performance of the model
and to reflect the application of ANN for value is 9% for this numerical example.
conceptual cost estimation in Ethiopian road
projects are percentage expected and mean
CONCLUSIONS
absolute percentage errors.
The research was conducted to provide the
The mean absolute percentage error of the road construction industry with a
conceptual cost estimation model calculated computerized and high-tech tool that can
for the 48 exemplars is 32.58%. This is a very estimate cost during the early phase of project
encouraging result with the provided set of development. The conclusions drawn from this
exemplars. The model is performed with only research are summarized as follows:
48 exemplars which is not nearly enough to 1. Neural Networks have been used in
represent road projects in Ethiopia. It is, construction engineering and
therefore, predicted to improve with the management for prediction and
provision of sufficient data to support all state- decision-making globally. However
of-affairs of each input parameter identified to advanced technological tools applied
affect total project cost as a representative worldwide that can expand
sample. The ANN model developed on this construction management roles are not
research cannot be readily used in the industry fully exhausted and exploited by the
but this research is an indication to the Ethiopian road construction industry.
prospect of application of ANN for cost 2. Based on the data analysis and
estimation of Ethiopian road projects during discussion the influential parameters
early phase of the project development. for conceptual stage include
27
Journal of EEA, Vol. 35, October, 2017
 construction duration, location, intensiveness and terrain classification
wearing surface class, project scope, should be further studied.
terrain classification, structural As mentioned above advanced cost
intensiveness, road length and year of estimation tool can also be applied to
construction. Ethiopian building construction projects. For
3. The mean absolute percentage error of instance the Ethiopian government is
the conceptual cost estimation model currently working towards provision
developed is 32.58%, which has of condominium apartments to the vast
proven that Neural Network is a urban population. Therefore, future
promising tool for use in the initial researches can be made in developing cost
stages of Ethiopian road construction estimation models for construction of
projects, providing construction building project.
managers access to well-organized and
scientifically systemized historical REFERENCES
cost data of similar projects on local
basis. [1] Sirous A. and Michael P., “Highway
4. The application of this cost estimation project cost estimating and management, The U.S.
tool increases accuracy of the cost department of transportation Federal highway
estimates. administration,” The State of Montana
Department of Transportation in cooperation with
The U.S. Department of Transportation Federal
Significantly through the minimization of Highway Administration, February 2009
assumptions cost estimators have to draw due [2] Ethiopian Roads Authority, Report on
to the existence of low information availability “Assessment of 15 years performance of Road
during the initial phase. Sector Development Program,” ERA, Ministry of
Transport, The Federal Democratic Republic of
RECOMMENDATIONS Ethiopia, January 2013
[3] Yinges T., “Causes of Cost and Time
Overrun on ERA Road Construction Projects,”
The aggregate recommendations drawn are
ERA Road Research Center, Ethiopia, 2016
summarized below in a category of the party [4] Zinabu T., “Causes of Contractor Cost
that is aimed to be addressed. Overrun in Construction Projects: The Case of
Decision-Makers in Ethiopian Road Ethiopian Construction Sector,” International
Construction Industry Journal of Business and Economics Research,
Decision makers of the industry include road Vol. 4 Issue 4: p180-91, August 2015
authorities, financers and policy makers on [5] Ministry of Transportation and
road construction in Ethiopia. The applications Infrastructure Project Management Support
of advanced project cost estimation tools Services: “Project Cost Estimating Guidelines,”
currently in use by the developed world are Version 01.02: p20 September 30, 2013
[6] Jelen, F. and Black J., “Cost and
relatively at an infant stage in Ethiopian road
Optimization Engineering,” 2nd edition,
construction relatively. Therefore it is McGraw-Hill, New York, 1983
recommended that authorities embrace and
adapt such computerized cost estimation [7] Ibrahim M., “Conceptual Cost Estimate of
techniques to be applied during early phase of Road Construction Projects in Saudi Arabia,”
project development. Therefore, the provision Jordan Journal of Civil Engineering, Vol. 7, Hail
of supporting policies along with University, KSA, 2013
implementation and enforcement technique [8] Michael D., Dell I., “The Architect’s
establishments that will enable the use of Handbook of Professional Practice: Detailed Cost
reliable cost estimation tool is recommended. Estimating,” 2003
[9] Hasan J., M.Sc. Thesis on “Parametric
Cost Estimation of Road Projects Using Artificial
Baseline for Future Research Neural Networks,” Gaza Deanery of Graduate
Studies, Faculty of Engineering, Civil Engineering
The discarded input parameters in this Department, Construction Management, The
research due to unavailability of data which Islamic University, Gaza, 2013
are project delivery system, structural [10] Emad E., Lecture note on “Construction
Project Management,” Structural Engineering

28
Journal of EEA, Vol. 35, October, 2017
 Department, Faculty of Engineering, Mansoura [17] Megha J. and Pathak K., “Applications of
University,” November 2015 Artificial Neural Network in Construction
[11] Strategic Assessment and Estimating Engineering and Management,” International
Office (SAEO) in alignment with the goals of the Journal of Engineering Technology, Management
Statewide Program Management Group (SPMG): and Applied Sciences, Volume 2 Issue 3, August
“Cost Estimating Manual,” p11, Washington 2014
State, November 2008 [18] Carsten P. and Thorsteinn R., “An
[12] Shihunegn A., M.Sc. Thesis on “Testing Introduction to Artificial Neural Networks,”
Regression Models to Estimate Costs of Road University of Lund, Sweden, 1991
Construction Projects in Ethiopia” School of [19] M. Hajek, “Neural Networks,” Neural
Graduate Studies, Construction Technology and networks.doc, University of KwasZulu-Natal,
Management, Addis Ababa Institute of 2005
Technology, 2014 [20] Nardos T., M.Sc. Thesis on “Application
[13] Amr A., “Parametric Cost Estimating of of Artificial Neural Network for model
Highway Projects using Neural Networks,” development of Conceptual Cost Estimation of
Faculty of Engineering & Applied Sciences, St. Road Projects in Ethiopia,” Construction
John's Newfoundland, Canada, July 1997 Technology and Management, Addis Ababa
[14] Ibrahim M., “Early Cost Estimating for Institute of Technology, Addis Ababa University,
Road Construction Projects Using Multiple July 2016
Regression Techniques,” Australasian Journal of [21] Ethiopian Road Authority Manual, “Site
Construction Economics and Building; Vol. 11 Investigation, Physiography, climate, geology and
Issue 4, 2011 soil distributions categorization Manual,” 2013
[15] Stuart J. Russell and Peter N., “Artificial [22] Gemechu W., Presentation on “Challenges
Intelligence: A Modern Approach, Prentice Hall, and prospects of Ethiopia’s cement industry:
Englewood Cliffs,” New Jersey, 1995 Strategic perspective,” Construction review
[16] Megha J. and Pathak K., “Applications of online By Africa for Africa, April 28, 2015
Artificial Neural Network in Construction [23] Google Earth, Search: “Dangote, Derba,
Engineering and Management,” International National, Messebo, and Mugher cement factory in
Journal of Engineering Technology, Ethiopia,” Accessed on April-May, 2016, https:/
Management and Applied Sciences, Volume 2 /www.google.com/earth
Issue 3, August 2014

29
Journal of EEA, Vol. 35, October, 2017