Estimation of Construction Project Cost Based on GA-BPNN
Guangying Jin
*a
and Chunhui Yang
b
School of Maritime Economics and Management, Dalian Maritime University, Dalian, China
Keywords: Construction Cost Estimation, Backward Propagation, Genetic Algorithm, Construction Enterprises, Project
Management.
Abstract: Elevated uncertainty over recent years with the unprecedented crisis of COVID - 19 have led to greater
difficulties for construction enterprises to control project costs. Meanwhile, estimating the project cost is
sophisticated, especially considering about actual situation of their project management team. Thus, it is a
challenge for construction enterprises to implement advanced decision support tools to estimate their project
cost in the construction process. This paper proposes a model to help construction enterprises to estimate
project costs. The model was developed using an approach by incorporating analytic hierarchy process (AHP)
and backward propagation neural network (BPNN) optimized by the genetic algorithm (GA). Thus, it can
involve subjective influencing factors and objective influencing factors. Moreover, a case study was carried
out on actual projects by different construction enterprises located in Shandong province. The result shows
that the implementation of the model can help construction enterprises to estimate construction costs with an
average error of less than 8%. Consequently, the model can help managers estimate project costs taking into
account the impact of the project management team.
1 INTRODUCTION
In 2022, Chinese national investment in real estate
development decreased by 5.4% and the investment
in housing decreased by 4.5% to 5,180.4 billion RMB
from January to June through the China National
Bureau of Statistics. As housing investment
decreases, local Chinese construction companies will
face greater and broader competition in the market.
Since the pandemic, uncertainties in the project
management process have increased. Meanwhile,
construction companies usually face problems such as
increased costs and schedule delays (Abdel-Hamid,
M. et al., 2021). Thus, calculating and controlling the
cost of construction projects is more complex and
time-consuming than before (Ahiaga-Dagbui et al,
2014).
Construction project cost, known as the sum of all
costs of a construction project. Effective control of
project costs is the key to ensuring corporate profits
(Fazil, 2021). Therefore, construction project cost
estimation remains a key part of construction project
management (El-Kholy,2021).
Until now, the widely used method for cost
estimation of construction projects has been modeling
a
https://orcid.org/0000-0002-4459-8425
b
https://orcid.org/0000-0001-6441-9511
with Glodon software. However, the construction
industry is highly data intensive and generates a large
amount of data during the construction phase, the
value of which is far from being fully utilized
(Chakraborty, D.et al., 2020). The application of
various data mining techniques has started to make
people aware of the value of data in the construction
industry. As a result, data mining techniques are
widely used for construction project cost estimation
(Elmousalami, 2020).
Hence, the main objective of this study is to
propose an estimation method considering
management factors and non-management factors to
help construction managers estimate their
construction project cost.
2 LITERATURE REVIEW
To study the application of data mining technology in
the field of construction cost, we conducted a search
of the relevant literature. The keywords searched were
“estimate construction cost’’, ‘‘predict construction
cost’’, and ‘‘construction cost management’’. As a
196
Jin, G. and Yang, C.
Estimation of Construction Project Cost Based on GA-BPNN.
DOI: 10.5220/0011917600003612
In Proceedings of the 3rd International Symposium on Automation, Information and Computing (ISAIC 2022), pages 196-201
ISBN: 978-989-758-622-4; ISSN: 2975-9463
Copyright
c
 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

result, the word cloud in the abstract of the searched
literature is shown in
Figure 1.
In Figure 1, we can find that construction cost
estimation has become an active field in the past
decade (Mahmoodzadeh, A.et al.,2021). Meanwhile,
the machine learning method is widely used in
construction cost estimation （Fan, M et al.，2021）
. In more detail, we see that support vector machine
techniques (SVM), case-based reasoning (CBR),
decision tree, and neural network (NN) have been
widely used.
Figure 1: Word cloud of abstract of the article searched.
The methods mentioned above are all based on
historical data, and hope to find certain reference
value for new construction projects. As we all known,
SVM performs well on small sample sets, CBR has a
certain dependence on the previous field in
prediction, and decision trees are prone to overfitting.
Considering these characteristics, this study adopts
the BPNN with strong fitting ability. In view of the
shortcomings of the BPNN, we use GA to optimize it.
To the best of our knowledge, most relevant
studies have focused on proposing construction cost
forecasting models based on historical construction
information. There are limited studies that consider
the construction reality of construction firms to
estimate the construction costs suitable for individual
firms. However, estimating the cost of a project is
important for project management (Rustamova, S.,
2021). To bridge the gap, this paper proposes a model
that considers the impact of management teams and
estimates project costs for each firm.
3 DESIGN CONSTRUCTION
COST ESTIMATION MODEL
3.1 Management and
Non-Management Influencing
Factors Selection and Evaluation
Both management-related factors and the
characteristics of the project itself will affect the
project cost, so we need to analyze these two factors.
Delphi method, also known as the expert opinion
method, is a method of expressing opinions
anonymously through multiple rounds of survey
experts' opinions on the issues raised in the
questionnaire. After repeated consultations,
summaries, and modifications, the final aggregate
into a consensus opinion of experts. In here, we use
the Delphi method and invite experts in project cost
management field to select the management factors
and non-management factors that have a greater
impact on the cost.
At the beginning, we invited experts to select the
influencing factors related to management and
obtained
Table 1. In Table 1, we can find that the
management factor (F1) includes three sub-factors
(SF1.1---SF1.3). Among them, SF1.1 is the subjective
score obtained through AHP, and SF1.2 and SF1.3 is
the situation of the on-site management team obtained
through investigation.
Table 1 Management-related influencing factors and
reasons for choice given by experts.
Management
influencing sub-
factors
(GA-BP model
input Variable)
Reasons given by review experts
in the field of construction.
Level of project
management team
(SF1.1)
The commitment of the site team
to this project and the level of
attention given to cost
management will directly affect
the cost.
Working years of
cost manager on
the project
(SF1.2)
Experienced cost managers know
which segments are prone to cost
overruns and can take preventive
measures.
The number of
professionals with
credentials on the
p
roject (SF1.3)
Qualified personnel with certain
project management knowledge
can better control and manage site
materials, personnel, etc.
Moreover, a complete project management team
consists of project general managers, business
managers, production managers, technical managers,
and so on. Each person has a different weight on the
Estimation of Construction Project Cost Based on GA-BPNN
197

team. AHP is one of the most widely used
multicriteria decision-making techniques (Luthra et
al., 2016). Thus, AHP is used to analyse the
importance of each project member.
In this research, experts were invited to compare
the importance of two pairwise members and to rate
the scale of importance of the chosen members by
using the values of importance 1-9 scale (Saaty,
1990). Through the evaluation of project management
experts, we get the judgment matrix. After that, we
normalize the judgment matrix in columns and
calculate the average of each row as the weight of
members. Calculate each person's weight score based
on how important each person is to cost management.
Further, we invited experts to select the non-
management influencing factors. We get the non-
management influencing factors in
Table 2. The non-
management factor (F2) includes eight sub-factors
(SF2.1---SF2.8). Because the BPNN only handles
digital inputs, we encode the character-type features,
and the encoding results are also shown in
Table 2.
3.2 GA-BPNN Design for Estimating
Construction Costs
In this paper, BPNN is used to find a function to fit
cost and impact factors. Thus, the input variables are
the cost influencing factors and the output variable is
the project cost. We show this in
Figure 2. The training
process of BPNN is divided into three steps:
Step (1): Forward Propagation.
Forward propagation proceeds in the direction from
the input layer to the output layer. This process
presents in Equation (1).
𝑆

=𝑤

𝑥

+𝑏

(1)


Where: 𝑠

is the input of activation function, m is the
number of layer, 𝑤

is the weight between node i and
node j, 𝑥

is the value of node i， 𝑏

is the threshold.
Then the node output is calculated by activation
function, such as Equation (2).
𝑥

=𝑓𝑠

 (2)
Where: 𝑥

is the output value of the node.
Table 2 Content of non-management factors and coding of
their characteristics.
Non-management
sub-factors
(Some other input
variables of GA-
BPNN)
GA-BPNN
input Variable ’s
Category
Co
de
The type of
foundation of building
(SF2.1)
Raft foundation 1
Independent foundation 2
Pile foundation 3
The type of Structure
of building (SF2.2)
Shear structure 1
Frame structure 2
Frame-shear structure 3
The number of above
ground floor (SF2.3)
Low-rise residential 1
Multi-store
y
residential 2
Mi
d
-rise residential 3
Hi
g
h-rise residential 4
The number of below
ground floor (SF2.4)
No basement 1
1 Floor for basement 2
2 Floors for basement 3
The building area of
building (SF2.5)
Less than 2500 𝑚

1
2500 𝑚

-5000 𝑚

2
5000 𝑚

-7500 𝑚

3
7500 𝑚

-10000 𝑚

4
More than 10000 𝑚

5
The type of interior
wall decoration
(SF2.6)
Plastered interior wall 1
Paint interior wall 2
Paint and plastered mix
wall
3
The type of exterior
wall decoration
(SF2.7)
Paint exterior wall 1
Brick exterior wall 2
Paint and veneer brick
mix wall
3
The type of ground
decoration (SF2.8)
Installed underfloor
heatin
g
1
Not installed underfloor
heatin
g
2
In here, we choose the Sigmoid function as the
activation function, such as Equation (3).
𝑥

=
1
1+𝑒


(3)
Step (2): Error Calculation.
The error between the predicted value and the real
value is calculated as Equation (4).
𝐸
(
𝑤,𝑏
)
=
1
2
𝑜

−𝑦





(
4
)
Where: 𝑜

is the output value, 𝑦

is the real value.
ISAIC 2022 - International Symposium on Automation, Information and Computing
198

Figure 2: The structure of cost estimation BPNN model and the input and output variables.
Step (3): Error Back Propagation.
The backward propagation proceeds in the direction
from the output layer to the input layer. In here, using
the gradient descent algorithm as the learning
algorithm of the network. The weight update formula
is Equation (5).
𝛥𝑤
(
𝑖,𝑗
)
=−𝜂
𝜕𝐸
(
𝑤,𝑏
)
𝜕𝑤
(
𝑖,𝑗
)
(5)
Where:
𝜂 is learning rate.
Figure 3: The structure of cost estimation BPNN model and the input and output variables.
Estimation of Construction Project Cost Based on GA-BPNN
199

Figure 4: The work process of GA-BPNN model in
estimating project cost.
When taking derivative layer by layer, follow the
chain derivative Equation (6).
𝜕𝐸
(
𝑤,𝑏
)
𝜕𝑤
(
𝑖,𝑗
)
=
𝜕𝐸
(
𝑤,𝑏
)
𝜕𝑜𝑢𝑡
𝜕𝑜𝑢𝑡
𝜕𝑛𝑒𝑡
𝜕𝑛𝑒𝑡
𝜕𝑤(𝑖,𝑗)
(6)
Where: 𝜕𝑜𝑢𝑡 is the output value of the previous level.
Finally, the updated weight 𝑤

is as Equation (7).
𝑤

=𝑤+𝛥𝑤
(
𝑖,𝑗
) (
7
)
The network often cannot guarantee that the
randomly set weights and thresholds can obtain the
best results. GA is used to optimize the initial weights
and thresholds of BPNN. As shown in
Figure 3.
3.3 The Work Process of Construction
Cost Estimation
In this study, we would like to explore the feasibility
of using models to predict project costs while also
considering managerial factors. The process design of
the project cost estimation is shown in
Figure 4.
4 CASE STUDY
As mentioned earlier, to train, validate and test the
GA-BPNN model. we have utilized a database that
we assembled based on civil construction projects
located in Shandong province (China) as the research
data.
We train the GA-BPNN model with the dataset.
The fit of the model is high during the training
process, and the 𝑅

of multiple training is above 0.92,
which indicates that the constructed network can
better best fit the parameters. Thus, we save the cost
estimation model. To test the estimation effect of the
model, we used another 15 projects to compare the
estimation results. We calculate the average error rate
to be within 8%. And the results are shown in F
igure
5.
Figure 5: Comparison of estimated and real values of GA-BPNN model and calculation of error rate.
ISAIC 2022 - International Symposium on Automation, Information and Computing
200

Meanwhile, we compared the prediction results of
other machine learning models. Each model is run 20
times and averaged. To assess the degree of
applicability of different models to this problem, we
counted the MAE, RMSE, and 𝑅

for each
prediction.
Table 3: Comparison of prediction performance of different
machine learning models.
NO. Model MAE RMSE
𝑅

1 Random Forest 177.42 250.78 0.73
2 SVM 94.68 152.67 0.91
3 BPNN 102.31 160.72 0.88
4 GA-BPNN 54.35 89.29 0.94
In Table 3, we can find that the GA-BPNN
performs best on our data set, followed by SVM, and
the random forest algorithm performs worst. These
results show the applicability of the GA-BPNN model
in estimating the cost of construction project.
5 CONCLUSION
This paper presents comprehensive descriptions of
the proposed GA-BPNN model and its application in
project cost estimation for construction enterprises.
Meanwhile, we considered the influence of
engineering project management factors and used the
Delphi method to effectively select the factors that
have a large impact on the project cost. And the
weights of different members were calculated by
AHP for project team level.
After simulation, the error is within the allowable
range, which has certain guiding significance for
managers to estimate the project cost according to
their own project conditions. Thus, it maybe be
feasible to consider the actual characteristics of the
enterprise management team in the process of project
cost estimation. Meanwhile, GA-BP model might be
reliable in solving the problem of project cost
estimation.
ACKNOWLEDGEMENTS
The study was supported by the Talent Research
Start-up Founding of Dalian Maritime University,
authorization code: 02502329.
REFERENCES
Abdel-Hamid, M., & Abdelhaleem, H. M. (2021). Project
cost control using five dimensions building information
modelling. International Journal of Construction
Management, 1-10.
Ahiaga-Dagbui, D. D., & Smith, S. D. (2014). Rethinking
construction cost overruns: cognition, learning and
estimation. Journal of financial management of
property and construction.
Chakraborty, D., Elhegazy, H., Elzarka, H., & Gutierrez, L.
(2020). A novel construction cost prediction model
using hybrid natural and light gradient
boosting. Advanced Engineering Informatics, 46,
101201.
El-Kholy, A. M. (2021). Exploring the best ANN model
based on four paradigms to predict delay and cost
overrun percentages of highway projects. International
Journal of Construction Management, 21(7), 694-712.
Elmousalami, H. H. (2020). Data on field canals
improvement projects for cost prediction using artificial
intelligence. Data in brief, 31, 105688.
Fan, M., & Sharma, A. (2021). Design and implementation
of construction cost prediction model based on SVM
and LSSVM in industries 4.0. International Journal of
Intelligent Computing and Cybernetics.
Fazil, M. W., Lee, C. K., & Tamyez, P. F. M. (2021). Cost
estimation performance in the Construction Projects: A
systematic review and Future Directions. International
Journal of Industrial Management, 11, 217-234.
Luthra, S., Mangla, S. K., Xu, L., & Diabat, A. (2016).
Using AHP to evaluate barriers in adopting sustainable
consumption and production initiatives in a supply
chain. International Journal of Production
Economics, 181, 342-349.
Mahmoodzadeh, A., Mohammadi, M., Daraei, A., Farid
Hama Ali, H., Ismail Abdullah, A., & Kameran Al-
Salihi, N. (2021). Forecasting tunnel geology,
construction time and costs using machine learning
methods. Neural Computing and Applications, 33(1),
321-348.
Rustamova, S. (2021). Application of PMI PMBOK
standard in the development of the Construction cost
management projects plan. International journal of
trends in business administration, 11(1).
Saaty, T. L. (1990). The analytic hierarchy process in
conflict management. International Journal of Conflict
Management.
Estimation of Construction Project Cost Based on GA-BPNN
201