Ijciet 06 10_006

http://www.iaeme.com/IJCIET/index.asp 62 editor@iaeme.com
International Journal of Civil Engineering and Technology (IJCIET)
Volume 6, Issue 10, Oct 2015, pp. 62-76, Article ID: IJCIET_06_10_006
Available online at
http://www.iaeme.com/IJCIET/issues.asp?JType=IJCIET&VType=6&IType=10
ISSN Print: 0976-6308 and ISSN Online: 0976-6316
© IAEME Publication
DEVELOPMENT OF THE ARTIFICIAL
NEURAL NETWORK MODEL FOR
PREDICTION OF IRAQI EXPRESS WAYS
CONSTRUCTION COST
Dr. Tareq Abdul Majeed Khaleel
Building and Construction Engineering Department
University of Technology, Iraq
ABSTRACT
The main objective of this research is to introduce a new and alternative
approach of using a neural network for cost estimation of the expressway
project at the early stage.
A preliminary literature survey and data collection have identified the
problem and led to the formulation of the research hypothesis that there is a
weakness in estimating the cost of the expressway construction projects
because the current available techniques are poor and suffer some
disadvantages such as being traditional, aged, slow and uncertain. Besides,
the need for a modern efficient construction cost estimation techniques that
have more advantages such as being modern, fast, accurate, flexible and easy
to use is of value. Also, the application of Artificial Neural Networks, as a
modern technique, in Iraqi construction industry is necessary to ensure
successful management, and many of the construction companies feel the need
of such system in project management
One model was built for the prediction the cost of expressway project. The
data used in this model was collected from Stat Commission for Roads and
Bridges in Iraq. It was found that ANNs have the ability to predict the Total
Cost for expressway project with a good degree of accuracy of the coefficient
of correlation (R) was 90.0%, and average accuracy percentage 89%.The
ANNs model developed to study the impact of the internal network parameters
on model performance indicated that ANNs performance was relatively
insensitive to the number of hidden layer nodes, momentum term, and learning
rate.
Key word: Cost Estimate, Expressway Project, Average Accuracy and
Artificial Neural Network

Development of The Artificial Neural Network Model For Prediction of Iraqi Expressways
Construction Cost
Cite this Article: Dr. Tareq Abdul Majeed Khaleel. Development of The
Artificial Neural Network Model For Prediction of Iraqi Expressways
Construction Cost. International Journal of Civil Engineering and
Technology, 6(10), 2015, pp. 62-76.
http://www.iaeme.com/IJCIET/issues.asp?JType=IJCIET&VType=6&IType=10
1. INTRODUCTION
Expressways were an expressway especially planned for high-speed traffic, usually
having few if any intersections, limited points of access or exit, and a divider between
lanes for traffic moving in opposite directions.
Conceptual cost estimate is one of the most important activities to be performed
during the project planning phase. It includes the determination of the project’s total
costs based only on general early concepts of the project (Kan, 2002). Like all other
planning activities, conceptual cost estimating is a challenging task. This is due to the
availability of limited information at the early stages of a project where many factors
affecting the project costs are still unknown
Major difficulties which arise while conducting cost estimation during the
conceptual phase are lack of preliminary information, lack of database of road works
costs, and lack of up to date cost estimation methods. Additional difficulties arise due
to larger uncertainties as result of engineering solutions, socio-economical, and
environmental issues. Parametric cost estimation or estimation based on historic
database during the conceptual estimate phase is widely used in developed countries.
However, developing countries face difficulties related to the creation of a road work
costs database, which may be used for cost estimation in either the conceptual stage or
the feasibility study of a project cycle.
One of the earliest papers to introduce the benefits and the implementation of
ANN in the civil engineering community is published by (Flood and Kartam,1994).
This research has opened the door for many proposals that suggest ML as the
preferred method to tackle various challenges in the construction industry. Wilmot
and Mei,(2005) introduced an ANN model for expressway construction costs. This
research used the following factors as a base for cost estimation: price of labour, price
of material, price of equipment, pay item quantity, contract duration, contract
location, quarter in which the contract was let, annual bid volume, bid volume
variance, number of plan changes, and changes in standards or specifications. The
main contribution of this work was that it covered all required factors. Nevertheless,
the validation of the proposed method and the data collection process used for training
and testing the results were not fully presented. Furthermore, (Hola and Schabowicz,
2010) developed an ANN model for determining earthworks’ execution times and
costs. Basically, this model was developed on the basis of a database created from
several studies that were carried out during large-scale earthwork operations on the
construction site of one of the largest chemical plants in Central Europe. However, the
validation of the presented results is not mentioned. Petroutsatou et al.,(2012)
introduced the ANN as a technique for early cost estimation of road tunnel
construction. The data collection strategy of this research was based on structured
questionnaires from different tunnel construction sites. The main drawback of this
research was the ignoring of some of the construction cost factors. Jafarzadeh et
al.,(2014) proposed the ANN method for predicting seismic retrofit construction
costs. This study selected data from 158 earthquake-prone schools. The validation of
this method is not clear. Recently, (Al-Zwainy.,2008) used the multi-layer

perceptron trainings using the back-propagation algorithm neural network is
formulated and presented for estimation of the total cost of highway construction
projects. twenty influencing factors are utilized for productivity forecasting by ANN
model. four model was built for the prediction the productivity of marble finishing
works for floors. It was found that ANNs have the ability to predict the total cost of
highway construction projects with a very good degree of accuracy of the coefficient
of correlation (R), and average accuracy percentage .
They concluded that neural networks performed the best prediction accuracy but
case – based reasoning indicated better results in long run. Accurate cost estimation at
the early stages of project development is not only a problem for developed countries
but also developing countries. Therefore, there is need for better cost estimation
techniques at the conceptual phase to be developed. The application of ANN systems
is growing rapidly in the financial and manufacturing sectors. Neural network systems
offer several advantages over traditional methods for the prediction of construction
projects' cost and duration. . (Boussabaine,1996).
ARTIFICIAL NEURAL NETWORK: BACKGROUND
According to Rumelhart et al. (1986), there are eight components of a parallel
distributed processing model such as the neural network. These eight components are
the processing units or neurons, the activation function, the output function, the
connectivity pattern, the propagation rule, the activation rule, the learning rule and the
environment in which the system operates. Neural networks are a series of
interconnected artificial neurons which are trained using available data to understand
the underlying pattern. They consist of a series of layers with a number of processing
elements within each layer. The layers can be divided into input layer, hidden layer
and output layer. Information is provided to the network through the input layer, the
hidden layer processes the information by applying and adjusting the weights and
biases and the output layer gives the output (Karna and Breen 1989). Each layer may
have a number of processing units called neurons. The inputs are weighted to
determine the amount of influence it has on the output (Karna and Breen 1989), input
signals with larger weights influence the neurons to a higher extend. An activation
function is then applied to the weighted inputs, to produce an output signal by
transforming the input. The input can be a single node or it may be multiple nodes
depicting different parameters where each of the input nodes acts as an input to the
hidden layer. The hidden layer consists of a number of neurons/nodes which calculate
the weighted sum of the input data.
Figure (1) shows how neural network adjusts the weights and biases by comparing
the output with the target. The weights are not fixed but they change over time by
gaining experience after several iterations (Rumelhart et al. 1986). Artificial neural
networks are used in pattern classification, clustering/categorizing, function
approximation, predicting, optimization, control and content-addressable memory
(Jain et al. 1996).

Construction Cost
Figure 1 Correction of error using target data (Demuth, 2006)
Back-propagation algorithm is simple and effective in solving large and difficult
problems (Alavala, 2008). Thus, it is used in learning process of the model. It
consists of two phases: forward pass and backward pass (Beale and Jackson, 1990).
In forward pass, the parameters of the input variables pass though the functions of the
network and an output data is produced, in the end. In backward pass, firstly the error
is calculated by subtracting the actual output from desired output. Then, it is
propagated backward through the network. The weights are adjusted during the
backward pass (Ayed, 1997). This process optimizes the weight parameters of the
model, decrease the error value and increase the prediction power of the ANN model.
There are methods that significantly improve the back-propagation algorithm’s
performance (Haykin, 1999):
 Sequential versus batch update: When the training data set is large and highly
redundant, sequential mode of back propagation learning could be preferred than
the batch mode of the algorithm.
 Maximizing information content: The training data should be strong enough to
maximize the learning rate or the model. There are two ways to form such strong
training information; using data that is having the largest training error, and
using data that is oppositely different the other data used before.
 Activation function: Using sigmoid activation function increases the learning
ability of the model. Applying hyperbolic tangent, a nonlinear sigmoid
antisymmetric activation function of sigmoid nonlinearity, is popular in this way.
 Learning from hints: Learning from a set of training examples deals with an
unknown input-output mapping function.
 Learning rates: Learning rate values are important for the network in learning
process. Neurons with many inputs should have smaller learning rate parameter,
or vice versa.
2. BACKGROUND OF EXPRESSWAYS COST PREDICTION
The variations of several parameters that influence a construction project costs create
complexities for developing an accurate model of future Expressways construction
costs determination. However, there have been numerous publications describing
methods and techniques that approximate the future projects‟ costs. Thus, the aim of
this research is to develop a sufficient and accurate method of forecasting the costs of
the future Expressways ‟ construction. The compiled data have been initially analyzed
based on Artificial Neural Networks (ANN).
Input
Neural network computes
weights
Compare with
target
Adjust weights computes
weights

3. IDENTIFICATION OF ANN MODEL VARIABLES
One of the most important tasks of this objective is to determine which variables are
important indicators. Once the appropriate variables have been determined, the cost
estimation can be performed either using a neural network or any other tool, such as
regression analysis.
This research describes the development of neural network models of total cost
structure work of Expressways project based on recent historical projects data. The
initial impetus for the research was the paucity of data available that can provide
reliable information about the costs. The data used to develop the neural network
model of estimation of the cost were past expressway contract data from Iraq from
2010 to 2014.
The model input variables for this model were consisting of six variables (i.e. V1,
V2, V3, V4, and V5and V6). There are two types of variables that might affect the
estimation of expressway project construction cost objective variables and subjective
variables
 Objective variables: This type comprised eleven variables, as the following
V1 Length of the Pavement in (km)
V2 Capacity –the number of standard Width lanes
V3 Interchanges- number of expressway interchanges
V4 Number of Stream Crossing
 b) Subjective variables: This type comprised nine variables, as the following
V5 Material- this classifies pavement as flexible and rigid. And assigns the values of 1 and 2
respectively to them
V6 Furnishing- Expressway furnishing level; without (1), normal (2), high standards (3).
3. DEVELOPMENT OF ANN MODEL
In an effort to develop a more realistic cost model, this study attempts to overcome
some of Neural Network drawbacks , and presents it as a simple and transparent
approach for use in construction. Accordingly, a three-layer Neural Network has been
simulated on a (NEUFRAME, Version 4) program that is easy to use, transparent, and
customary to many practitioners in construction. The simulation of Neural Networks
on a NEUFRAME program presents its underlying mathematical formulas in a simple
and fully controllable form. . Figure (2) shows the scheme of the NEUFRAME 4
program which is built to determine the relationship between the independent
variables (inputs) and the dependent variable (output).
Artificial neural network models need to be in a systematic manner to improve its
performance. Such Method needs to address major factors such as, development of
model inputs, data division and pre-processing, development of model architecture,
model optimization (training), stopping criteria, and model validation, (Shahin et al,
2002) . These factors are explained and discussed below.

Construction Cost
Figure 2 Graphing Component of NEUFRAM 4 Program
4. MODEL INPUTS AND OUTPUTS
The selection of model input variables that have the most significant impact on the
model performance is an important step in developing ANN models. Presenting as
large a number of input variables as possible to ANN models usually increases
network size, resulting in a decrease in processing speed and a reduction in the
efficiency of the network, (Shahin, 2003)
It is generally accepted that eleven parameters have the most significant impact on
the cost estimation of Expressways projects, and are thus used as the ANN model
inputs. These include Objective variables and Subjective variables
The output of the model is the cost of Expressways project. A code is used in this
research to identify the names of the different models developed. The code consists of
two parts separated by a hyphen. The first part represents an abbreviation of the
current output (i.e. Total Cost Expressways, ID). The second part denotes the model
number. Hence, for example “TCE –1” represents Total Cost model number one.
5. PRE-PROCESSING AND DATA DIVISION
Data processing is very important in using neural networks successfully. It determines
what information is presented to create the model during the training phase. It can be
in the form of data scaling, normalization and transformation. Transforming the
output data into some known forms (e.g. log., exponential, etc.) may be helpful to
improve ANN performance. Thus, the logarithm of total cost Expressways is taken
before introducing forward in the next steps.
The next step in the development of ANN models is the division of the available
data into their subsets, training, testing and validation sets. trail–and-error process was
used to select the best division, by using Neuframe software. The network that
performs best with respect to testing error was used in this work (compared with other
criteria to evaluate the prediction performance, training error and correlation of
validation set). Using the default parameters of the software, a number of networks
with different divisions were developed and the results are summarized in table (1) It
can be seen that the best data subsets division is (80-10-10) % according to lowest
testing and training error coupled with highest correlation coefficient of validation set
(90.30%). Thus, this division was used in this model

Table 1 Effect of data division on performance of ANNs
Data Division training
error%
testing
error%
coefficient
correlation(r)%Training% Testing% Querying%
65 20 15 8.90 9.00 77.35
60 20 20 7.23 8.40 70.88
60 15 25 7.85 7.57 76.59
65 15 20 7.40 7.48 68.64
50 30 20 7.36 7.44 77.64
70 15 15 7.56 7.35 79.64
65 10 25 7.55 7.21 81.26
70 12 20 6.75 6.98 81.42
70 15 20 7.18 6.98 81.52
80 10 10 6.14 5.94 90.30
The effects of using different choices for divisions (i.e. striped, blocked, and
random) were investigated and it was shown in table (2), it can be seen that the
performance of ANNs model was relatively insensitive to the method of division. The
better performance was obtained when the striped division was used, according to
lowest testing (5.94%) and training error (6.14%) coupled with highest correlation
coefficient of validation set (90.30%).
Table 2 Effects of method division on ANNs performance
Data Division% choices
of division
training
error%
testing
error%
coefficient
correlation(r)%Training Testing Querying
80 10 10 Striped 6.14 5.94 90.30
80 10 10 Blocked 9.99 8.98 77.90
80 10 10 Random 9.09 8.88 75.50
6. SCALING OF DATA
Once the available data have been divided into their subsets, the input and output
variables are pre-processed by scaling them to eliminate their dimension and to ensure
that all variables receive equal attention during training. Scaling has to be
commensurate with the limits of the transfer functions used in the hidden and output
layers (i.e. –1.0 to 1.0 for tanh transfer function and 0.0 to 1.0 for sigmoid transfer
function). The simple linear mapping of the variables,
extremes to the neural
network’
s practical extremes is adopted for scaling, as it is the most commonly used
method, (Shahin, 2003). As part of this method, for each variable x with minimum
and maximum values of xmin and xmax, respectively, the scaled value xn is calculated
as follows:
minmax
min
n
xx
xx
x



(1)

Construction Cost
7. MODEL ARCHITECTURE, OPTIMIZATION AND STOPPING
CRITERIA
One of the most important and difficult tasks in the development of ANN models is
the determination of the model architecture. The network that performs best with
respect to the lowest testing error followed by training error and high correlation
coefficient of validation set was retrained with different combinations of momentum
terms, learning rates and transfer functions in an attempt to improve model
performance. Consequently, the model that has the optimum momentum term,
learning rate and transfer function was retrained a number of times with different
initial weights until no further improvement occurred.
The network of (Model 2) is set to one hidden layer with default parameters of
software (learning rate equals to 0.2 and momentum term equals to 0.8). A number of
trials were carried out with one hidden layer and 1, 2, 3…, 21 hidden layer nodes
(2I+1) (where I the number of input nodes) and the results are graphically in figure.
(3). It can be seen that the two hidden nodes have the lowest prediction error. Thus, it
was chosen in this model.
Figure 3 Performance of ANNs model with different hidden nodes (Model 2)
6.0%
6.5%
7.0%
7.5%
8.0%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
No. of Nodes
TrainingError
6.0%
6.5%
7.0%
7.5%
8.0%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
No. of Nodes
TestingError
72.5%
77.5%
82.5%
87.5%
92.5%
97.5%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
No. of Nodes
CorrelationCoeff.(r)

Figure (3) shows that the network with two hidden node has the lowest prediction
error for the testing test (5.60%). Therefore, two hidden node was chosen in this
model. It is believed that the network with two hidden node is considered optimal.
The effect of the momentum term on model performance was investigated for the
model with two hidden nodes (learning rate = 0.20). The results are summarized in
table (3). It can be seen that the optimum value for momentum term is 0.8, which has
the lowest prediction error; hence it was used in this model.
Table 3 Effects Momentum Term on ANNs performance (Model 2)
Parameters
Effect
Momentum
Term
training
error%
testing
error%
coefficient
correlation(r)%
Model No.
(TCSW-2)
choices of division
(Striped)
Learning Rate
(0.2)
No. of Nodes
(2)
Transfer function in
hidden layer
(Sigmoid)
output layer
(Sigmoid)
0.1 7.69 5.74 85.44
0.2 7.59 5.74 85.66
0.3 7.49 5.73 86.55
0.4 7.48 5.73 86.85
0.5 7.48 5.74 86.95
0.6 7.38 5.75 86.77
0.7 7.29 5.77 87.99
0.8 7.20 5.60 90.55
0.9 7.54 5.82 88.44
In addition, the effect of the learning rate on the model performance was
investigated (momentum term = 0.8) for Model 2. The results are summarized in table
(4). the optimum value for learning rate is 0.2, which have lowest prediction error,
hence it was used in this model.
Table 4 Effects Learning Rate on ANNs performance (Model 2)
Parameters
Effect
Learning
Rate
training
error%
testing
error%
coefficient
correlation(r)%
Model No.
(TCE-2)
choices of division
(Striped)
Momentum Term
(0.8)
No. of Nodes
(2)
hidden layer
(Sigmoid)
output layer
(Sigmoid)
0.1 6.98 6.83 87.97
0.2 7.20 5.60 90.55
0.3 7.42 5.86 88.78
0.4 7.44 5.87 88.76
0.5 7.45 5.91 89.22
0.6 7.49 5.94 89.76
0.7 7.46 6.99 89.57
0.8 7.46 6.99 89.12
0.9 7.65 7.00 89.20

Construction Cost
The effects of using different transfer functions (i.e. sigmoid and tanh) were
investigated and it was shown in table (5), it can be seen that the performance of
ANNs model was relatively insensitive to the type of the transfer function. The better
performance was obtained when the tanh transfer function was used for hidden and
output layers, which have lowest prediction error coupled with highest correlation
coefficient (r).
Table 5 Effects of transfer function on ANNs performance (Model 2)
Parameters
Effect
Transfer Function
training
error%
testing
error%
coefficient
correlation(r)%
Hidden
Layer
Output
Layer
Model No.
(TCE-2)
choices of division
(Striped)
No. of Nodes
(2)
Momentum Term
(0.8)
Learning Rate
(0.2)
sigmoid sigmoid 7.20 5.60 90.55
tanh tanh 7.88 7.68 88.66
sigmoid tanh 7.84 7.74 85.53
tanh sigmoid 7.88 7.58 80.03
8. ANNS MODEL EQUATION (MODEL 2)
The small number of connection weights obtained by Neuframe for the optimal
ANNs model (Model TCE-2) enables the network to be translated into relatively
simple formula. The structure of the ANNs model is shown in figure. (7), while as
connection weights and threshold levels (bias) are summarized in table (10).
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Figure 7 Structure of the ANNs optimal model (TCE-2)
TCE-2
1
2
3
4
5
6
7
8
X1
X2
X3
X4
X5
X6
9

Table 10 Weights and threshold levels for the ANNs optimal model (Model TCSW-2)
Hidden
layer
nodes
wji (weight from node i in the input layer to node j in the hidden layer) Hidden
layer
threshold θj
i=1 i=2 i=3 i=4 i=5
j=7
0.433 0.169 -0.300 -0.400 -0.500
0.11i=6
-0.211
j=8
i=1 i=2 i=3 i=4 i=5
0.39
-0.800 -0.199 -0.500 -0.511 -0.200
i=6
0.333
Output
layer
nodes
wji (weight from node i in the hidden layer to node j in the output layer) Output
layer
threshold θji=7 i=8
j=9 -0.10 -0.835 0.31
Using the connection weights and the threshold levels shown in Table (10), the
predicted of total cost can be expressed as follows:
)tanh0.835tanh10.031.0( 21
1
1
xx
e
TCE 


(2)
Where:
X1= {θ7+ (w7-1*V1)+(w7-2*V2)+(w7-3*V5)+(w7-4*V7)+(w7-5*V7)+(w7-6*V8) } (3)
X2= {θ8+ (w8-1*V1)+(w8-2*V2)+(w8-3*V5)+(w8-4*V7)+(w8-5*V7)+(w8-6*V8) } (4)
It should be noted that, before using Equation 3 and 4, all input variables (i.e. V1,
V2, V3, V4, V5 and V6), need to be scaled between 0.0 and 1.0 using Equation (1)
and the data ranges in the ANN model training (see Table 7). It should also be noted
that the predicted value of total cost obtained from Equation 6.14 is scaled between
0.0 and 1.0 and in order to obtain the actual value this total cost has to be re-scaled
using Equation (1) and the data ranges in Table (7) The procedure for scaling and
substituting the values of the weights and threshold levels from Table (10), Equations
(2) and (3) and (4) can be rewritten as follows:

Construction Cost
55.1
1
11.4
)tanh0.835tanh10.031.0( 21


  xx
e
TCE
(5)
And
X1={17.088+(0.1*V1)+(0.03*V2)+(-0.01*V3)+(-0.005*V4)+(0.01*V5)+(-0.01*V6)} (6)
X2={46.8954+(-0.13*V1)+(-0.016*V2)+(-0.11*V3)+(-0.012*V4)+(0.01*V5)+(-0.03*V6)}
(7)
9. VALIDITY OF THE ANN MODEL (MODEL TC-1)
The statistical measures used to measure the performance of the models included:
Where:
 Mean Absolute Percentage Error (MAPE),
n
A
EA
MAPE
n
i







 
 1
%100*
(8)
 Average Accuracy Percentage (AA %)
MAPEAA  %100% (9)
 The Coefficient of Determination (R2
);
 The Coefficient of Correlation (R);
The results of the comparative study are given in Table (11). The MAPE and
Average Accuracy Percentage generated by ANN model (TCE-2) were found to be
11% and 89% respectively. Therefore it can be concluded that ANN model (Model 2)
shows a good agreement with the actual measurements.
Table 11 Results of the Comparative Study
Description ANN for Model TC-1
MAPE 11%
AA % 89%
R 90.0%
R2
81.0%
To assess the validity of the ANNs model for the total cost of expressways project
(TCE), the logarithm of predicted values of TCE are plotted against the logarithm of
measured (observed) values of TCE for validation data set, as shown in figure (8). It
is clear from figure (8). The generalization capability of ANNs techniques using the
validation data set. The coefficient of determination (R 2
) is (81.0%), therefore it can
be concluded that ANNs model (Model 2) show very good agreement with the actual
measurements.

Figure 8 Comparison of predicted and observed total cost structure work for validation data
10. CONCLUSION AND FUTURE RESEARCH
Through the development of expressways construction cost estimation model
using neural network and the development of the proposed program, the following
points are concluded:
 a) Neural networks have demonstrated to be a promising tool for use in the
conceptual stages of construction projects when typically only a limited or incomplete
data set is available for cost analysis.
 b) In this study, one hidden layer with two hidden node for model (TCE-2) is
practically enough for the neural network analysis. The findings show that one ANN
model is able to learn the cause-effect relationships between input and output, during
the training stage, and obtained Average Accuracy percentage (AA) of 89% and the
coefficient of correlation (R) was 90.0%
Finally, future research directions are suggested for cost estimation in order to
overcome the gaps that have been discussed. These directions are as follows.
 Providing cost estimation proposals that encourage the acquisition of human
expertise: however, this releases the construction cost estimation from human
dependability. Computerized expert systems are the better mechanism that might be
used to replace human expertise.
 Developing several ANN models to demonstrate the ways in which different types of
civil engineering problems ensure the successful development and application of this
technology to civil engine erring problems.
 A research may be done on applying the same techniques to develop managements
systems for production rates of any constructions operations such as: earthmoving,
concreting etc.
R2
= 0.8105
0
1
2
3
4
5
6
0 1 2 3 4 5 6
Logarithm of Observed Total Cost
LogarithmofPredictedTotalCost
R2
=0.81

Construction Cost
REFERENCES
[1] Alavala, R.C. (2008), Fuzzy Logic and Neural Networks: Basic Concepts and
Applications, Retrieved December 22, 2010,
fromhttp://site.ebrary.com.libezproxy.iyte.edu.tr:81/lib/iyte/docDetail.action?docI
D=10318726&p00=neural%20computing%3A%20an%20introduction
[2] Al-Zwainy, Faiq, M. S. (2008), The Use of Artificial Neural Network for
Estimating Total Cost of Highway Construction Projects, a thesis submitted to the
Civil Engineering Department, College of Engineering, Baghdad University, Ph.
D
[3] Ayed A.S. (1997), Parametric Coat Estimating of Highway Projects using Neural
Networks, A thesis submitted to the School of Graduate Studies in the partial
fulfillment of the requirements for the degree of Master of Engineering, Faculty
of Engineering and Applied Sciences. Canada: National Library of Canada.
Retrieved December 09, 2010, from
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.152.7423&rep=rep1&t
ype=pdf
[4] Beale, R., & Jackson, T. (1990), Neural Computing: An Introduction, Bristol:
Institute of Physics Publishing
[5] Boussabaine, A. H. (1996), the Use of Artificial Neural Networks in Construction
Management: A Review, Journal of Construction Management and Economics,
14, pp. 427-436.
[6] Demuth, H., Beale, M., and Hagan, M. (2006), Neural network toolbox User’s
guide, Version 5, Natick, Massachusetts.
[7] Flood I. and Kartam N, Neural networks in civil engineering. I: principles and
understanding, Journal of Computing in Civil Engineering, 8(2), pp. 131–148,
1994.
[8] Haykin, S. (1999), Neural Networks: A Comprehensive Foundation, Upper
Saddle River, NJ: Prentice Hall International Inc.
[9] Hola, B. and Schabowicz, K., Estimation of earthworks execution time cost by
means of artificial neural networks, Automation in Construction, 19(5), pp. 570–
579, 2010.
[10] Jafarzadeh R, Ingham, S. Wilkinson, V. Gonz´alez, and A. A. Aghakouchak,
Application of artificial neural network methodology for predicting seismic
retrofit construction costs, Journal of Construction Engineering and
Management, 140(2), Article ID04013044, 2014.
[11] Jain, A. K., Mao, J., and Mohiuddin, K. M. (1996), artificial neural networks: A
tutorial, Computer, 29(3), 31-44.
[12] Kan, P. (2002), Parametric Cost Estimating Model for Conceptual Cost
Estimating of Building Construction Projects, Ph.D. Thesis. University of Texas,
Austin, USA
[13] Mohanraj A, Manoj Prabhakar S and Rajendran M Predicting the Strength of
Self-Compacting Self-Curing Concrete Using Artificial Neural Network.
International Journal of Civil Engineering and Technology, 5(12), 2013, pp. 239
- 247.
[14] Karna, K. N., and Breen, D. M. (1989), An artificial neural networks tutorial:
Part 1–basics, The international journal of neural networks, 1(1), 4-23.
[15] Petroutsatou K., Georgopoulos E., Lambropoulos S., and Pantouvakis J. P., Early
cost estimating of road tunnel construction using neural networks, Journal of
Construction Engineering and Management, 138(6), pp. 679–687, 2012.

[16] Rumelhart, D. E., McClelland, J. L and the PDP research group (1986),
ParallelDistribution Processing: Exploration in the microstructure of cognition,
1:Foundations, The MIT Press, Massachusetts.
[17] Shahin, M.A., (2003), Use of Artificial Neural Networks for Predicting
Settlement of Shallow Foundations on Cohesionless Soils, Ph.D. Thesis,
Department of Civil and Environmental Eng., University of Adelaide.
[18] Wilmot C. G. and Mei B, Neural network modeling of highway construction
costs, Journal of Construction Engineering and Management, 131(7), pp. 765–
771, 2005.
[19] Hyeong-Joo Kim, Jose Leo Mission and Jeong-Hee Ko. Development Artificial
Neural Network (Ann) Prediction of One-Dimensional Consolidation in
Homogeneous Clay under Uniform Applied Load. International Journal of Civil
Engineering and Technology, 4(4), 2013, pp. 135 - 145.
[20] Mohanraj A, Manoj Prabhakar S and Rajendran M Predicting the Strength of
Self-Compacting Self-Curing Concrete Using Artificial Neural Network.
International Journal of Civil Engineering and Technology, 5(12), 2013, pp. 239
- 247.

Ijciet 06 10_006

Recommended

Recommended

More Related Content

What's hot

What's hot (18)

Viewers also liked

Viewers also liked (11)

Similar to Ijciet 06 10_006

Similar to Ijciet 06 10_006 (20)

More from IAEME Publication

More from IAEME Publication (20)

Recently uploaded

Recently uploaded (20)

Ijciet 06 10_006