SSP - JOURNAL OF CIVIL ENGINEERING Vol.

18, Issue 1, 2023

DOI: 10.2478/sspjce-2023-0005

Improvement of Productivity in Buildings Construction

Chemseddine Dehchar1*, Khaled Boudjellal1, Mohamed Bouabaz1

1
Faculty of Technology, Department of Civil Engineering, LMGHU Laboratory, University of 20 August 1955,
Skikda, Algeria
*e-mail: [email protected] / [email protected]

Abstract

Improving productivity in construction projects has long been a major concern, and much research has been carried
out to try to ameliorate construction productivity. To this end, this study aims to improve and increase the
productivity rate of flat slab formwork used in residential construction projects. A survey consisting of 150
questionnaires was undertaken to identify the factors that influence on the productivity. Based on the relative
Importance Index (RII), data on eleven factors deemed to affect productivity were selected. A collection of 100
data points from various sites were utilized to develop two models. Firstly, an Artificial Neural Network (ANN)
model was employed, and secondly, a parametric approach was investigated. The data were divided into two sets,
with 70% of the data used for training and the remaining 30% used for testing. The models' performance was
evaluated using the Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE) values. In the test
phase, the artificial neural network model yielded an MSE value of 2.6610e-4 and a MAPE value of 4.9227,
whereas the parametric model produced an MSE of 0.040 and a MAPE of 9.525. It was found that the artificial
neural network model provided reliable prediction accuracy compared to the parametric model. However, the
artificial neural network approach can be selected as a robust model in predicting and controlling the productivity
rate in local construction projects by using the developed model based on the identified factors.
Keywords: Artificial Intelligence, Construction Industry, Artificial Neural Network, Labour Productivity,
Parametric Approach

1 Introduction

In the construction industry, productivity can be defined as the quantity of production work per
corresponding input [1]. This relationship can take different forms, defining labour productivity
when using work-hour as an input or cost productivity when cost input is taken into account [2,
3]; however, this simple definition of factors does not seem realistic and cannot give robust and
accurate results, as omitting the influence of other factors can affect productivity. Various
research studies have been conducted over many years to estimate, predict and improve the
productivity rate in the construction industry, taking into account the effect of different
influencing factors.
The review of the literature in the area of construction and specifically productivity
improvement can be divided into two main parts, the first focuses on the identification of factors
affecting productivity, and the second deals with applications of artificial intelligent.

1
Dehchar C., Boudjellal K., and Bouabaz M.

The pace of labour productivity is a critical factor that directly influences the outcome of
projects, ultimately determining their success or failure [4, 5]. According to [6], factors
affecting labour productivity may vary from country to country and from site to site, and may
also vary within a site. [7] Stated that, labour productivity is influenced by various factors,
including external conditions, site conditions, and workers' characteristics. The combined
influence of these factors determines the overall level of efficiency and effectiveness attained
in construction tasks. [8] Invited 180 Kuwaiti companies to participate in a sample survey that
included 45 factors grouped into four groups, to identify the factors most influencing Kuwaiti
projects, the result of the statistical analysis determined that the total project cost could be
reduced by focusing on the design phase. The influence of several factors on labour productivity
in Malaysian construction sector was examined in [9], the authors found that project
management skills were the factors that had the greatest impact on productivity, and also stated
that the implementation of new technologies in construction sector has positive effects on
labour productivity. Another study by [10] highlights that the management group ranks first in
three categories, which includes 30 factors studied for their effects on labour productivity in
Egyptian construction firms, the authors also found that workers experience and skills were the
most important factor. A questionnaire survey conducted in Trinidad and Tobago by [11], to
identify the influencing factors on construction projects, were investigated on 30 contractors
who are members of the Contractors Association. The survey included 42 factors categorized
into four groups: Management, Technological, Human/Labour, and External. Both close-ended
and open-ended queries were used, and the respondents were asked to score each factor based
on a Likert Scale, ranging from 1 to 4, indicating the effect level. To determine the rank of each
factor, the researchers calculated the Relative Importance Index (RII%) and ranked the factors
accordingly. The top three influencing factors identified were: the lack of labour supervision,
unrealistic scheduling and expectation of labour performance, shortage of experienced labour.
In [12], 52 factors, were used to investigate their influence on productivity in Yemen, by
calculating the relative importance index, the authors found that workers experience and skills
were the most significant factors. [13] Proposed a fuzzy fault tree-based approach to identify
factors influencing labour productivity in Lithuania. Initially, they gathered data from 15
experts who evaluated 27 factors using a Likert scale, ranging from 1 to 5, to indicate the degree
of severity. After this step, 18 factors with a Relative Importance Index (RII) greater than 0.7
were selected for the subsequent analysis. In the next phase, the authors constructed a fault tree
structure through multiple sessions and expert interviews to define the interrelations among the
selected factors. Finally, the fuzzy approach was employed to evaluate the contribution of each
factor. The results of their analysis indicated that two factors, namely "inflation in the cost of
execution" and "improper project financing," were identified as the most influencing factors
impacting labour productivity in Lithuania. Another study conducted by [14] aimed to identify
the factors affecting labour productivity in the South Africa construction industry based on
contractors' perceptions. A questionnaire survey consisting of 41 factors was randomly
distributed to 96 contractors. The obtained responses were then analysed using descriptive and
inferential statistics. The factors were ranked according to their P-values, and the most
significant influencing factors on labour productivity were found to be: excessive bureaucracy,
late delivery of materials, industrial action resulting from political activities. Authors in [15]
identified factors affecting labour productivity in Australian construction projects by following
a three-step approach. In the first step, they conducted a questionnaire to gather data on various
factors related to labour productivity in construction projects. The second step involved drawing

2
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

a cause-and-effect feedback loop to identify the complicated interrelated links between the 38
factors. Lastly, the authors used the Decision-Making Trial and Evaluation Laboratory
(DEMATEL) method to prioritize the factors. The results of the study revealed that project size,
level of skill and experience, and communication problems with foreign workers were the most
significant factors affecting labour productivity in Australian multi-storey building construction
projects. In another research conducted by [16], authors examined the influence of weather
conditions on labour productivity in the United Kingdom. The study involved collecting data
over a span of 5 years, from September 2017 to August 2022. To gather information, the
researchers administered 28 online questionnaire surveys to relevant participants. Based on
their findings, the authors proposed implementing modular techniques in construction, wherein
certain elements are produced in factory settings with controlled environments. This approach
is expected to mitigate the adverse effects of weather on construction projects, consequently
reducing overall costs.
In the second half of the 1980s, Artificial Neural Networks were developed and used to solve
various construction problems such as cost estimation, duration and productivity. Numerous
models were produced [17–19],and are still in progress to this day. [20] Examined the influence
of ten factors, on Turkish formworkers' productivity. A three-layer feed-forward back
propagation neural network was used, using these factors as input variables, a hidden layer
containing five hidden neurons and an output neuron to estimate the labourers' productivity rate,
compared to the estimate of the Turkish Ministry of Public Works and Settlement, the
simulation results indicate that the model developed accurately predicted the total man-hours
required. In another approach, [21], also formulated a multi-layer ANN with the
backpropagation algorithm using ten influencing factors, as input variables to assess the labour
productivity rate for finishing work on marble floors. The architecture of the model used was a
three-layer neural network with one hidden node in the hidden layer and used the hyperbolic
tangent sigmoid function as a transfer function, a degree of accuracy of 90.9% was recorded,
which proved that Network, the Backpropagation Neural Network, the Radial Base Function
Neural Network and the Adaptive Neuro-Fuzzy Inference System, and compared their results
to choose the best model the developed model can reliably predict the productivity rate, it was
also found that the most influential factors on productivity were age, number of workers and
experience. Two ANNs models with different architectures were developed by [22], using three
categories of influencing factors, the capabilities of a Feed-Forward neural network and a
Radial-Based NN to predict masonry crew productivity were compared; the results showed that
the RBNN technique predicted productivity better than the FFNN. Authors in [23] took thirteen
factors that were found to have a negative effect on labour productivity and considered them as
inputs for developing an ANN model to estimate bricklayer productivity, a gradient descent
momentum backpropagation was used as a training algorithm for the model which has three
hidden layers and a hyperbolic tangent sigmoid transfer function; accurate results were obtained
using this architecture, in addition, the results of the sensitivity analyses led to the conclusion
that the wall thickness was the factor that most influenced the labour productivity rate. [24]
Developed four ANN models based on different techniques, including the General Regression
Neural for predicting the productivity rate of formwork labours, the performance results show
that the Backpropagation NN is the best model among the four techniques developed in the
study.Authors in[25]stated that labour productivity in Saudi construction projects is affected by
a number of factors and can be quantified using mathematical models. For this purpose, the
author developed four different models based on multilayer perceptron neural network

3
Dehchar C., Boudjellal K., and Bouabaz M.

(MLPNN), general regression neural network (GRNN), support vector machine (SVM), and
multiple additive regression trees (MART) to assess the productivity rate of concrete
construction activities and compared the results of the developed models. The GRNN model
was found to be the best estimator model for labour productivity of steel fixing and concrete
pouring, while the MART model provided better results than the others for formwork assembly
productivity. In order to ensure effective and efficient project delivery,[26] utilized an artificial
neural network (ANN) to assess the relationships between various factors and bricklaying
productivity during the pre-planning phase. The developed model included 7 input factors, 5
hidden neurons, and one output. The results showed good performance, indicating that the ANN
model can be utilized to enhance project delivery by providing a more realistic estimation of
productivity rates. This, in turn, leads to better planning, cost estimation, and resource
allocation. In a study conducted by [27], authors used sets of five Artificial Neural Network
models to develop a single model capable of estimating steel reinforcement works. They
collected 145 data points from construction sites in Poland, which were then divided into 90%
for training and 10% for testing purposes. The proposed model demonstrated satisfactory
performance, indicating its robustness in estimating steel reinforcement works. [28] Developed
an Artificial Neural Network (ANN) model to assess the productivity rate of piping assembly
activities in Brazilian projects. They tested 108 different Feedforward Neural Network (FFNN)
architectures, each with 14 input neurons, to select the best model. The research results
indicated that the Mean Square Error (MSE) of the chosen model reached the value of 9.67E-
04, demonstrating satisfactory performance and accuracy. A hybrid model combining artificial
neural network (ANN) and Grasshopper Optimisation Algorithm (GOA) was developed by[29].
The objective was to identify the factors with the greatest impact on labour productivity in
Iranian construction projects and improve the accuracy of labour productivity prediction. Out
of the 19 factors initially considered, only 6 factors were identified as influential and utilized to
develop the predictive ANN model. [30] Employed Artificial Neural Networks (ANN) to
predict the productivity rate for brick masonry work. The authors conducted a questionnaire
survey that initially included forty-four factors. However, based on the Relative Importance
Index (RII) analysis, only the top 13 ranking factors were considered as inputs for the model.
The developed model utilized a three-layer backpropagation-feedforward neural network with
one hidden layer consisting of 30 hidden neurons. The Mean Squared Error (MSE) was used as
a measure of accuracy for the model. The results indicated that the developed model effectively
estimated brick masonry productivity by considering the impact of the selected 13 factors.
Parametric model was developed in the early 1960s, it has been used to find the relationship
between the response and one or more factors, usually by applying regression analysis to the
parameters [31]. [32] Developed a parametric equation based on multiple linear regression
(MLR) analysis to model the labour productivity of marble floor finishing works in Iraq.100
data including 10 influencing factors were collected and used to predict productivity, good
performance, and a high correlation was obtained, indicating a good relationship between the
response and the factors, and the model can be reliably used to estimate productivity. In another
study conducted by [33] to estimate the percentage loss or increase in construction labour
productivity, 14 influencing factors were defined and used to develop a regression model, the
author argued that the proposed model can accurately predict construction labour productivity.
[34] Examined the influence of various buildability factors on the labour productivity of
structural elements, by developing a MLR model to assess the relationship between design
characteristics and labour productivity. The results show that the model can provide information

4
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

for designers to adapt their designs to achieve the best labour productivity performance.Authors
in [35] conducted a study to examine the impact of worker experience and skill level on masonry
construction productivity in Vietnam. They developed a logistic regression model to estimate
productivity and found that there was no difference in average productivity between different
worker experience groups. However, they did observe a significant difference in productivity
between worker skill groups. [36] Employed a multiple linear regression method to analyse the
impact of three factors on labour productivity in lightweight brick wall installation and
lightweight brick wall plastering in Indonesia. The researchers achieved an impressive accuracy
of 99.43% and 99.04% for the two respective models, indicating the robustness and reliability
of the developed models. In their study conducted at eight construction sites in Niš, Serbia, [37]
Focused on analysing concreting processes, particularly those related to columns and walls.
They gathered 60 data points to develop regression analysis and Simulation models aimed at
improving the accuracy of forecasting productivity rates for concrete works. The findings of
the study indicated that these proposed models have the potential to enhance decision-making
and improve the accuracy of planning in construction projects involving residential buildings
in Niš, Serbia.
The Algerian construction industry is currently facing a major problem in accurately measuring
construction labor productivity. This issue is mainly attributed to the continued utilisation of
traditional technologies, which has significantly hindered progress in the construction sector.
The challenges are particularly evident in the execution of flat slab formwork activities,
resulting in excessive resource consumption in terms of manpower, time, and cost. To enhance
construction efficiency, time management, and cost control, it is crucial to address the
productivity issues related to flat slab formwork. By optimizing the flat slab formwork
productivity in Algerian building projects, it becomes possible to achieve faster construction
cycles, lower labor costs, and improve overall project performance. This, in turn, leads to more
efficient resource utilization and successful project outcomes.
While previous studies have examined construction productivity in a general context, however,
there appears to be a noticeable gap in the existing research regarding comprehensive
investigations into the factors influencing productivity, particularly in Algerian construction
projects. Additionally, there is a lack of intelligent methods that specifically concentrate on
accurately measuring and estimating labor productivity rates for flat slab formwork within the
context of the local construction industry. Since our study focuses on a local problem with
specific characteristics compared to other world regions, we are faced with the following issues:
• What are the key factors that influence productivity in Algerian construction projects?
• How do the identified factors influence the productivity rate of flat slab formwork in
Algerian construction projects?
• How can we estimate this productivity?
This paper aims to address two primary objectives: identifying the key factors that affect
productivity in Algerian construction projects and developing intelligent models capable of
estimating the labour productivity rate of flat slab formwork. By conducting research on this
topic, the study intends to fill the existing knowledge gap and contribute valuable insights to
both academia and industry practitioners. The findings of this research will provide valuable
guidance for productivity improvement and decision-making in the management of
construction projects in Algeria.

5
Dehchar C., Boudjellal K., and Bouabaz M.

2 Methodology

This study aims to improve the labour productivity rate of flat slab formwork in Algerian
building projects. To achieve the objective mentioned, the methodology presented in this
section was adopted.

2.1 Questionnaire Survey Analysis

Identifying factors that negatively affect labour productivity in construction projects has been
a major concern for managers [32, 38]. After careful analysis of the top influential factors cited
by researchers [10, 12], [21-23], [39–43], the factors that appeared repeatedly were identified
and extracted. Moreover, by considering on-site observations of construction projects, a
questionnaire survey was designed including 16influencing factors classified into three main
groups as shown in Table 1.The survey consisted of closed-ended questions that were sent by
email to local expert, who were asked to rate the impact of each factor on labour productivity
using a five-point Likert scale, ranging from 1 (indicating the least impact) to 5 (indicating the
greatest impact). Respondents were specifically asked to indicate the extent to which a specific
factor affected labour productivity according to their own perception. The designed
questionnaire survey was divided into three sections:
1 The first shows the respondents' personal information (position and professional experience)
2 The second section consists of three groups with 16 influencing factors, the respondents rank
the importance of each of them based on a five-point Likert scale, where 1 indicates the least
impact while 5 represents the greatest impact.
3 The last section represents a space dedicated to the respondents' opinions.
Table 1: Questionnaire survey factors
Questions (Factors)
Group 01: Management Group 02: Factors related to workers Group 03: External factors
Small crew size
Low wages Weather condition
Formwork condition Work-hour/absenteeism Humidity
Delay in payment of the Worker’s experience High temperature
company by the project owner Quantity of installed work Low temperature
Congestion in the workspace Worker’s age Shortage of building
Poor control materials
Low workers motivation

2.1.1 Sample Size

For a more diverse and precise analysis, we specifically targeted project managers, technical
managers, site managers, civil engineers, and architects working in the public and private
sectors to gather their perceptions and viewpoints on the various factors that influence labour
productivity in the Algerian construction industry. A statically representative sample of the
population was obtained using a formulae developed by [12, 44], as expressed bellow:

6
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

𝑚
𝑛= (𝑚−1) (1)
1+ 𝑁

where, n is the sample size of the limited population, m is the sample size of the unlimited
population, and N is the sample size of the available population. m is estimated as follows:
𝑧 2 × 𝑝(1 − 𝑝)
𝑚= (2)
𝜀2
z represents the statistic value for the confidence level used (z = 1.645 for 90% confidence
level), ε is the sampling error of the point estimate, and p is the value of the population
proportion that is being estimated, for which [45], proposed using a conservative value of p =
0.50 to obtain a sample size at least as large as needed. The value of m can be obtained than:
(1.645)2 × 0.50(1 − 0.50)
𝑚= = 67.65 ≃ 68 (3)
0. 12
In this study, a total number of 150 available samples were selected, based on this; the required
representative sample size of the population is obtained as follows:
68
𝑛= (68−1)
≃ 47 (4)
1+ 150

A total of 150 samples were distributed by email and only 56 feedbacks were received, this
number of completed questionnaires met and exceeded the required sample size. The
respondents in the samples are project managers, technical managers, site managers, civil
engineers, and architects, working in Algerian construction industry. 48.22% of them having
an experience of 1 to 4 years, while 44.64% have an experience of 05 to 09 years, and the
remaining 7.14% have 10 or more years of experience.

2.1.2 Defining the Influencing Factors

The Relative Importance Index (RII) technique was used to define the most influential factors;
this index was calculated according to equation (6).
∑5𝑖=1 𝑛𝑖 × 𝑥𝑖
𝑅𝐼𝐼(%) = ( ) × 100 (6)
5 ∑5𝑖=1 𝑥𝑖
where, ni indicates the Likert scale from 1 to 5, and xi denotes the frequency of each ni.
The results of the RII show that three factors had an RII greater than 80%, whereas nine factors
have an RII between 70 and 80%, while the RII of the other four factors is in the range of 60%
and 70%. Table 2 summarizes the results of RII and the rank of each factor.
Table 2: RII (%) Results and rank of each factor
Factors RII (%) Rank
Small crew size 78.21 07
Low wages 89.64 01
Formwork condition 84.29 03
Congestion in the workspace. 71.79 12

7
Dehchar C., Boudjellal K., and Bouabaz M.

Poor control 74.64 09

Delay in payment of the company by the project owner 74.29 10
Low workers motivation 86.07 02
Work-hour /absenteeism 78.93 04
Worker’s experience 78.57 05
Quantity of installed work 78.57 05
Worker’s age 68.57 13
Weather condition 68.57 13
Humidity 65.00 16
High temperature 75.36 08
Low temperature 67.14 15
Shortage of building materials 73.93 11
In this study, factors with an RII (%) greater than 75% were taken into account. On this basis,
the most relevant factors deemed to affect the productivity rate are: workers' wages, workers'
motivation, formwork condition, number of working hours, workers' experience, crew size,
quantity of installed work, and temperature. Table 3 summarises all the factors considered.
Table 3: The selected influential factors
Factors Categories Classification
Number of working hours Quantitative Numerical
Number of formworkers Quantitative Numerical
Number of unskilled workers Quantitative Numerical
Minimum workers' experience Quantitative Numerical
Maximum workers' experience Quantitative Numerical
Mediocre
Formwork condition Qualitative Average
Good
Extra money
Type of workers' motivation Qualitative Recuperation days
Piecework
Temperature Quantitative Numerical
Wage of formworkers Quantitative Numerical
Wage of unskilled workers Quantitative Numerical
Surface area of the slab formed Quantitative Numerical

2.2 Data Collection

The data for this study were collected daily, based on in site observations on the flat slab
formwork task, a number of 100 data samples, comprising 11 quantitative and qualitative
influencing factors, were compiled from construction projects in different regions of Algeria,
the qualitative factors used in this study were transformed into numerical values for use in the
models (Table 4). As indicated in the previous sections, productivity can be calculated in
different ways (total or partial factor productivity) [46], the type of labour productivity, which
is of interest in this study, is defined as the ratio of output to worker-hour (Equation 7), noting
that the output presents the surface area of the slab that was carried out in square metres.
output
productivity rate = (7)
man work hour

8
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

Table: 4 Transformation of qualitative factors into numerical form

Qualitative factors Numerical transformation
Formwork state Mediocre -0.50
Average 0.00
Good 0.50
Type of motivation Extra money 0.10
Recuperation days 0.15
Piecework 0.20

2.3 Artificial Neural Network Model Development

2.3.1 Data Preprocessing

The preprocessing phase consists of preparing the data before introducing them into the model,
which gives the network the capacity to converge more quickly and to better generalise the
results obtained. In this phase the input and target data were normalised to be within the range
of [0 to 1] both for the training and the test phase using equation (8).
𝑋𝑖 − 𝑋𝑖𝑚𝑖𝑛
𝑋𝑛 = (8)
𝑋𝑖𝑚𝑖𝑛 𝑖
𝑚𝑎𝑥

where: Xn: the value of the variable (xi) after normalization. And Xi: the value of the variable
before normalization. Xi min; Xi max are the minimum and maximum value of (Xi).

2.3.2 Data Processing

Is the process by which the collected data is organised and introduces into the ANN model, to
obtain better generalisation of the results, for this purpose, the designed set was divided into
training and test subsets, the training subset is used to initialise the weights and biases of the
network. According to [47], each network starts training from arbitrary initial weights and
biases, while the test subset is used to verify the network design, testing the efficiency of the
model on the new data. During the modelling phase, a phenomenon can occur when errors begin
to increase in front of the test data after providing very small values in the training sets, this
phenomenon is known as over fitting, in this case, the model adjusted its parameters well and
stored all the training courses, but when new data was introduced into the network, the model
failed to adapt and generalise the new situation. Different methods have been used to avoid this
phenomenon; in this study the Bayesıan regularisation technique was used.
Simulation software was utilized to create, train and test the networks, with random division of
the designed set into 70% training, and 30% into test sets.

2.3.3 Model Architecture and Adapted Algorithm

The model architecture represents the number of appropriate layers, and the way these layers
are connected, as well as the type of algorithm chosen. In this study, several architectures were
tested to obtain better accuracy, the model with three layers having 11 input neurons, two hidden
layers having 15 and 10 hidden neurons in the first and second hidden layers respectively, and

9
Dehchar C., Boudjellal K., and Bouabaz M.

one output neuron as presented in Figure 1. Backpropagation was used as a learning algorithm
with Bayesian regulation.

Figure 1: The architecture of the ANN model developed

2.3.1 Transfer Function

The transfer function is a monotonic, continuous, increasing and differentiable function applied
to the sum of the weighted inputs of the model to produce their resulting outputs. This function
consists of two sub-functions used by the neurons in the different layers. Firstly, the neuron in
each layer receives a linear combination of weighted inputs, determining the sum of the
information collected, using the activation function A(x) which is given as follows:
𝑁

𝐴𝑖 (𝑥) = ∑(𝑊𝑖𝑗 × 𝑋𝑗 ) − 𝜃𝑖 (9)

𝑗=1

where: Xj: the inputs, and Wij: the weights and θi is the threshold (bias).
In the next step, the neurons use an O(A) output function, to keep the output values within a
specified range, operating it on the activation function which has a scalar format and returns
the scalar output neurons. Thus, the transfer function O(A(x)) is the combination of the two
previous functions (activation and output function), which is used to produce the final output
neurons [48]. Different transfer functions are available in the Artificial Neural Network model
used to pass information from one layer to the other [47], the most commonly used are log-
sigmoid, tan-sigmoid, and linear (purelin). Each function of the latter has its own range that
generates output results within it. The logsig produces results between 0 and 1 while tansig
returns outputs in the range of -1 and 1, however, the specific range of purelin is: [-∞ + ∞].
The transfer functions adopted in this study for the implementation of the ANN model were the
log-sigmoid transfer function (logsig) and the hyperbolic tangent sigmoid (tansig) which were
used in the first and second hidden layers, respectively, while the pureline linear transfer
function was used in the output layer equation (10), equation (11) and equation (12).

10
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

1
𝐿 𝑜𝑔 𝑠 𝑖𝑔(𝐴) = (10)
(1 + 𝑒𝑥𝑝(−𝐴))

2
𝑇𝑎𝑛𝑠𝑖𝑔(𝐴) = −1 (11)
(1 + 𝑒𝑥𝑝(−2𝐴))

𝐹(𝐴) = 𝐴 (12)

2.3.2 Training and Testing Phase

Network training is the first process carried out by the model to select its parameters, it consists
of adjusting the values of weights and biases until a minimum error value is reached and the
best performance is obtained. The Bayesıan regularization training algorithm was used in this
study to obtain better results. This algorithm consists of adding an additional term (penalty
term), which represents the sum of the squares weights to the equation error, and then
propagating the resulting errors back to readjust the weights and biases, and minimizing the
errors to obtain the best prediction of the output.
Out of the 100 datasets collected, a number of 70 datasets were used for training in the first
phase, while the remaining 30 datasets were retained to test the model in order to generalize the
model to the new data, and to obtain better performance in predicting the results.
The Mean Square Error (MSE) and the Mean Absolute Percentage Error (MAPE) were used in
this study to calculate the performance of the model, as follows:
∑𝑛𝑖=1(𝑥𝑖 − 𝐸𝑖 )2
𝑀𝑆𝐸 = (13)
𝑛
𝑛
100 𝑥𝑖 − 𝐸𝑖
𝑀𝐴𝑃𝐸 = ×∑| | (14)
𝑛 𝑥𝑖
𝑖=1

where, n is the number of the data, E is the model outputs, and xi is the actual value.

2.4 Parametric Analysis

Parametric analysis is a method that links a variable to be estimated (labour productivity) and
the different factors that influence it through a mathematical relationship. The technique most
commonly used to develop the parametric equation is Multiple Linear Regression (MLR).

2.4.1 Development of Multiple Linear Regression Model

At this stage, 70% of the total data collected, were used to develop the model using Statistical
Product and Solutions Services (SPSS) software. The correlation test was carried out to check
the relationship between the response and the explanatory variables. The results of the
correlation coefficient (R) and determination coefficient (R2) presented in Table 5 show a
strong correlation between the response (flat slab formwork productivity rate) and the

11
Dehchar C., Boudjellal K., and Bouabaz M.

explanatory variables, indicating that they have a good relationship. A significant value for the
model was found to be less than 0.05, meaning that the regression line fits the data well.
Table 5: Model Results Summary
Coefficient of Coefficient of Significance of the
correlation (R) determination (R2) model
0.958 0.917 0.000

Table 6 summarises the results of MLR coefficients and the resulting equation for estimating
the productivity rate of flat slab formwork in construction projects takes the form below:
Y = 3, 239 − 0, 304X1 − 0, 088X 2 − 0,139X 3 − 0, 021X 4
+0, 007X 5 − 0, 015X 6 − 1, 356X 7 + 0, 011X8 (15)
−0, 004X 9 + 0, 000103 X10 − 0, 000053 X11

Table 6: Coefficients of Multiple Linear Regression

Model Unstandardized Coefficients
ß Std. Error
(Constant) 3.239 0.716
X1 -0.304 0.024
X2 -0.088 0.035
X3 -0.139 0.014
X4 -0.021 0.024
X5 0.007 0.010
X6 -0.015 0.047
X7 -1.356 1.302
X8 0.011 0.001
X9 -0.004 0.006
X10 0.000103 0.000046
X11 -0.000053 0.000039

2.4.2 Validation of the Developed Model

Model validation is the stage during which the new data was used on the explanatory variables
(factors) to test the ability of the parametric equation developed in equation (15), to estimate
and predict the productivity rate of the flat slab formwork. In this respect, a set of 30 data was
applied to the developed parametric equation, and the results of the predicted productivity rate
were compared with the actual productivity rate. The MSE and MAPE measures were
calculated to verify the performance of the model with the new data.

3 Results and Discussion

This paper aims to develop a machine learning model capable of predicting and estimating the
labour productivity rate for flat slab formwork assembly. In this respect, an Artificial Neural
Network was first trained using 70% of the data sets and tested using 30% of the data sets.
Various ANN architectures were tested with different combinations in the number of hidden

12
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

layers and hidden neurons as well as in the learning algorithm, and in the different ANN
parameters, to choose the best model that gives the best performance, the final structure adopted
in this study was a three-layer feed-forward back-propagation neural network having two
hidden layers with 15 hidden neurons and 10 hidden neurons in the first and second hidden
layers respectively. The Bayesian regularization technique was adopted as the learning
algorithm. The modelling phase consists of training the network first; in this step, the model
shows its ability to predict the results of the labour productivity rate with lower values of the
MSE and the MAPE, as shown in Figure 2. Testing the network is the next step carried out by
the model to examine its ability to handle unseen data. Good results from MSE and MAPE were
obtained at this stage as well, showing the model's ability to predict the labour productivity
rates of flat slab formwork, as shown in Figure 3. In line with the results from the MSE and
MAPE, it can be said that the model developed performs well and that no significant difference
could be observed between the predicted and actual labour productivity rates. Table 7
summarises the overall results of the measures. The regression analysis in the developed ANN
model was carried out to examine the relationship between actual targets and estimated output,
a correlation coefficient of 0.9994 was obtained indicating a good linear correlation between
forecasted and actual labour productivity (Figure 4).
In the second part of the study, a parametric analysis of the collected data was carried out by
developing a Multiple Linear Regression model. The results of the correlation coefficient and
the determination coefficient rather than the significant value of the developed equation show
a strong correlation between the response (labour productivity rate) and the explanatory
variables (factors). A good result from MSE and MAPE was obtained, indicating that the
parametric equation developed has the capacity to estimate and predict the labour productivity
rates of flat slab formwork (Figure 2). A set of 30 data was applied to the developed parametric
equation to validate it and the results of the predicted labour productivity rates were compared
to the actual labour productivity rates. Based on the MSE and MAPE results obtained during
the validation phase, it can be stated that the parametric equation developed works well and can
be used reliably to predict the labour productivity rates of flat slab formwork (Figure 3). Table
8 lists the results of the performance of the parametric model.
Finally, from the results of the two methods above, it can be seen that both the ANN model and
the parametric model can be reliably used to estimate and predict the labour productivity rate.
However, significant differences between the performances of the two models developed were
observed, so we can conclude that the ANN model gives better results and predicts the labour
productivity rates of flat slab formwork better than the parametric model.
Table 7: Results of the Artificial Neural Network performance
Performance
MSE MAPE %
Phase
Training phase 8.0796e-05 0.82832
Test phase 2.6610e-04 4.9227
Table 8: Results of the parametric equation performance
Performance
MSE MAPE %
Phase
Development phase 0.048 9.395
Validation phase 0.040 9.525

13
Dehchar C., Boudjellal K., and Bouabaz M.

1
Actuel Target
Predicted ANN outputs
0.9 Predicted parametric outputs

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 10 20 30 40 50 60 70

Figure 2: Predicted Neural Network and parametric outputs versus actual targets in the training phase
0.8
Actuel Target
Predicted ANN outputs
Predicted parametric outputs
0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30

Figure 3: Predicted Neural Network and parametric outputs versus actual targets in the test phase
All: R=0.99945
1
Data
0.9 Fit
Y = T
0.8
Output ~= 1*Target + -0.0045

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
Target

Figure 4: The regression analysis and correlation coefficient R in ANN model

14
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

Additionally, we investigated samples of previous studies conducted by many researchers on

formwork productivity modelling. Results were reported regarding the performance of each
model. For instance, [49] obtained a MAPE of 14.55% and 17.69% for the ANN and MLR
models, respectively, for column formwork productivity. In another study by [25], the MART
model was found to be the best among the four developed models, with an MSE of 0.0937.
Furthermore, [24] developed four models for slabs, walls, and columns formwork productivity,
it was found that Backpropagation Neural Network was the best model which achieved an MSE
of 0.018.
Table 9 describes a comparison study conducted on existing models versus the developed
models.
Table 9: Results comparison among existing models
Existing Research The used model Performances results
in the test phase
Developed research study Artificial Neural MSE = 2.6610e-04
Network
MAPE = 4.9227 %
Multiple Linear MSE = 0.040
Regression
MAPE = 9.525 %
S. Nassar and A. Khaleel. (2019) Artificial Neural MAPE = 14.55 %
Network

Multiple Linear MAPE = 17.69 %

Regression
E. A. Mlybari. (2020) MART MSE = 0.0937

Golnaraghi et al. (2019) Backpropagation MSE = 0.018

Artificial Neural
Network

By comparing the outcomes of the mentioned models with models developed in this study, it
can be observed that the latter exhibits significantly higher performance accuracy. The achieved
MSE and MAPE values suggest that our ANN and MLR models are better in predicting
formwork productivity and providing more accurate results. This confirms the reliability and
efficiency of the methods used in this study, which leading to a significant advancement in the
field of formwork productivity modelling.

4 Conclusion

This study presented a development of a multilayer Artificial Neural Network and a parametric
equation based on MLR models aimed at predicting and increasing the labour productivity rate
of the flat slab formwork in Algerian construction projects. A survey of 150 questionnaires was
distributed to define the most influential factors. The total number of data collected on the 11
influencing factors defined in this study was 100, which were then randomly divided into two
main subsets. An Artificial Neural Network model was first developed using 70% of the data
set, to train the network to update weights and biases, and the remaining 30% was used to test

15
Dehchar C., Boudjellal K., and Bouabaz M.

the network for better generalisation on the new data. The model architecture used was a three-
layer feed-forward ANN with a backpropagation algorithm and Bayesian regularisation as a
learning function. A parametric equation was developed in the second part of this study, also
using the same 70% of the dataset to establish a mathematical equation based on the multiple
linear regression method. The remaining 30% of the dataset was used to validate the equation
against the new data.
The results of this study clearly show that the ANN model developed is more efficient and gives
more accurate prediction results than the parametric equation. However, it can be noticed that
both developed models have a good relationship between the predicted results and the actual
targets. Moreover, a high performance was recorded, indicating that both models behave well
with the unseen data and produce results reasonably close to the actual targets. The results of
this research study allow concluding that the ANN model and the parametric equation
developed can be reliably used to predict the measurement of the productivity rate with the
incorporation of influencing factors deemed to affect the labour productivity of the flat slab
formwork task.

References

[1] Gerek, I. H., Erdis, E., Mistikoglu, G. & Usmen, M. A. (2014). Evaluation of plastering crew
performance in building projects using data envelopment analysis. Technological and Economic
Development of Economy, 22(6), 926–940.
[2] Thomas, H. R., Maloney, W. F., Horner, R. M. W., Smith, G. R., Handa, V. K. & Sanders, S. R.
(1990). Modeling construction labor productivity. Journal of Construction Engineering and
Management, 116(4), 705–726.
[3] Yi, W. & Chan, A. P. C. ( 2014). Critical review of labor productivity research in construction
journals. Journal of Management in Engineering, 30(2), 214–225.
[4] Boughaba, A. & Bouabaz, M. (2020). Identification and risk management related to construction
projects. Advances in Computational Design, 5(4), 445–465.
[5] Boulkenafet, N., Boudjellal, K. & Bouabaz, M. (2023). Optimization of productivity in the
rehabilitation of building linked to BIM. Advances in Computational Design, 8(2), 179–190.
[6] Olomolaiye, P. O., Jayawardane, A. K. W. & Harris, F. C. (1998). Construction Productivity
Management. England: Addison Wesley Longman.
[7] Florez-Perez, L. Song, Z. & Cortissoz, J. C. (2022). Predicting construction productivity with
machine learning approaches. Proceedings of the International Symposium on Automation and
Robotics in Construction. 2022-July, (107–114).
[8] Jarkas, A. M. & Bitar, C. G. (2012). Factors affecting construction labor productivity in Kuwait.
Journal of Construction Engineering and Management. 138(7), 811–820.
[9] Karim, N. A., Hassan, S., Yunus, J. & Hashim, M. (2013). Factors Influence Labour Productivity
and the Impacts on Construction Industry. Caspian Journal of Applied Sciences Research,
2(August), 349–354.
[10] El-Gohary, K. M. & Aziz, R. F. (2013). Factors influencing construction labor productivity in
Egypt. Journal of Management in Engineering, 30(1), 1–9.
[11] Hickson, B. G. & Ellis, L. A. (2014). Factors affecting Construction Labour Productivity in
Trinidad and Tobago. The Journal of the Association of Professional Engineers of Trinidad and
Tobago, 42(1), 4–11.
[12] Alaghbari, W., Al-Sakkaf, A. A. & Sultan. B., (2019). Factors affecting construction labour
productivity in Yemen. International Journal of Construction Management, 19(1), 79–91.
[13] Shoar, S. & Banaitis, A. (2018). Application of fuzzy fault tree analysis to identify factors
influencing construction labor productivity: A high-rise building case study. Journal of Civil

16
 SSP - JOURNAL OF CIVIL ENGINEERING Vol. 18, Issue 1, 2023

Engineering and Management, 25(1), 41–52.

[14] Adebowale, O. & Smallwood, J. (2020). Contractors’ perceptions of factors affecting
construction labour productivity in South Africa. Journal of Contemporary Managemen, 17(2),
326–346.
[15] Nasirzadeh, F., Rostamnezhad, M., Carmichael, D. G., Khosravi, A. & Aisbett, B. (2022).
Labour productivity in Australian building construction projects: a roadmap for improvement.
International Journal of Construction Management, 22(11), 2079–2088.
[16] Roberts, D., Park, K., Ankrah, N. & Kim, K. P. (2023). The effect on cost when utilising modular
construction techniques-The impact of weather on labour productivity. International Journal of
Sustainable Building Technology and Urban Development, 14(1), 77–95.
[17] Bouabaz, M. & Hamami, M. (2008). A cost estimation model for repair bridges based on artificial
neural network. American Journal of Applied Sciences, 5(4), 334–339.
[18] Otmani, A., Bouabaz, M. & Al-Hajj, A. (2020). Predicting Maintenance and Rehabilitation Cost
for Buildings Based on Artificial Neural Network and Fuzzy Logic. International Journal of
Computational Intelligence and Applications, 19(1), 1–13.
[19] Peško, I. et al. (2017). Estimation of costs and durations of construction of urban roads using
ANN and SVM. Complexity. 2450370.
[20] Umit Dikmen, S. & Sonmez, M. (2011). An artificial neural networks model for the estimation
of formwork labour. Journal of Civil Engineering and Management, 17(3), 340–347.
[21] Al-Zwainy, F. M. S., Rasheed, H. A. & Ibraheem, H. F. (2012). Development of the construction
productivity estimation model using artificial neural network for finishing works for floors with
marble. ARPN Journal of Engineering and Applied Sciences, 7(6), 714–722.
[22] Gerek, I. H., Erdis, E., Mistikoglu, G. & Usmen, M. (2015). Modelling masonry crew
productivity using two artificial neural network techniques. Journal of Civil Engineering and
Management, 21(2), 231–238.
[23] Aswed, G. K. (2016). Productivity Estimation Model for Bracklayer in Construction Projects
Using Neural Network. Al-Qadisiyah Journal For Engineering Sciences, 9(2), 183–199.
[24] Golnaraghi, S., Zangenehmadar, Z., Moselhi, O. & Alkass, S. (2019). Application of Artificial
Neural Network ( s ) in Predicting Formwork Labour Productivity. Advances in Civil
Engineering, 2019, 1–11.
[25] Mlybari, E. A. (2020). Application of soft computing techniques to predict construction labour
productivity in Saudi Arabia. International Journal of GEOMATE, 19(71), 203–210.
[26] Bokor, O., Florez-Perez, L., Pesce, G. & Gerami Seresht, N. (2021). Using Artificial Neural
Networks to Model Bricklaying Productivity. Proceedings of the 2021 European Conference on
Computing in Construction, 2021, July (52–58).
[27] Juszczyk, M. (2020). Analysis of labour efficiency supported by the ensembles of neural
networks on the example of steel reinforcement works. Archives of Civil Engineering, 66(1), 97–
111.
[28] Sarmento, M., Sc, M., Luiz, M., Ferreira, R. & Sc, D. (2021). Sizing an Artificial Neural Network
for Modelling Piping Assembly Labor Productivity Under Different Conditions. International
Journal of Engineering Research and Applications, 11(2), 16–22.
[29] Goodarzizad, PP., Mohammadi Golafshani, E. & Arashpour, M. (2021). Predicting the
construction labour productivity using artificial neural network and grasshopper optimisation
algorithm. International Journal of Construction Management, 1–28.
[30] Lawaju, N., Parajuli, N. & Shrestha, S. K. (2021). Analysis of Labor Productivity of Brick
Masonry Work in Building Construction in Kathmandu Valley. Journal of Advanced College of
Engineering and Management, 6, 159–175.
[31] Phaobunjong, K. (2002). Parametric Cost Estimating Model for Conceptual Cost Estimating of
Building Construction Projects. University of Texas at Austin.
[32] Al-Zwainy, F., Abed - Alla Eiada, A. & Abed - Almajed Khaleel, T. (2016). Application

17
Dehchar C., Boudjellal K., and Bouabaz M.

Intelligent Predicting Technologies in Construction Productivity. American Journal of

Engineering and Technology Management, 1(3), 39–48.
[33] El-Kholy, A. M. (2010). Productivity performance estimation in Egyptian construction projects
based on regression analysis. Mansoura Engineering Journal (MEJ), 35(01), 44–54.
[34] Narayanan, K. M. A. & Aswathi, M. P. (2018). Labour Productivity Model for Structural
Elements By Varying Buildability Factors using Multiple Regression Analysis. International
Journal of Advance Research, Ideas and Innovations in Technology, 4(2), 139–146.
[35] Khanh, H. D., Kim, S. Y., Van Khoa, N. & Tu, N. T. (2021). The relationship between workers’
experience and productivity: a case study of brick masonry construction. International Journal
of Construction Management, 1–10.
[36] Paikun, P., Fatimah, C., Nugroho, N. S. & Kusmana, D. (2021). Labor Productivity Rating Model
for Light Brick Wall Installation in Residential Projects. ASTONJADRO, 10(2), 260–270.
[37] Matejević-Nikolić, B. & Živković, L. (2022). Comparative Representation of Two Models for
Predicting the Productivity of Column and Wall Concreting Process. Buildings, 12(11), 1809.
[38] Al-Zwainy, F. M. S., Abdulmajeed, M. H. & Aljumaily, H. S. M. (2013). Using Multivariable
Linear Regression Technique for Modeling Productivity Construction in Iraq. Open Journal of
Civil Engineering, 03(03), 127–135.
[39] Enshassi, A., Mohamed, S., Mustafa, Z. A. & Mayer, P. E. (2007). Factors affecting labour
productivity in building projects in the Gaza strip. Journal of Civil Engineering and
Management. 13(4), 245–254.
[40] Fagbenle Olabosipo, I., Ogunde Ayodeji, O. & Owolabi James, D. (2011). Factors Affecting the
Performance of Labour in Nigerian Construction Sites. Mediterranean Journal of Social
Sciences, 2(2), 251–257.
[41] El-Gohary, K. M., Aziz, R. F. & Abdel-Khalek, H. A. (2017). Engineering Approach Using ANN
to Improve and Predict Construction Labor Productivity under Different Influences. Journal of
Construction Engineering and Management, 143(8), 1–10.
[42] Hafez, S. M. (2014). Critical Factors Affecting Construction Labor Productivity in Egypt.
American Journal of Civil Engineering, 2(2), 35.
[43] Kaming, P. F., Olomolaiye, P. O., Holt, G. D. & Harris, F. C. (1997). Factors influencing
craftsmen’s productivity in Indonesia. International Journal of Project Management, 15(1), 21–
30.
[44] Hogg, R. V. & Tanis, E. a. (2009). Instructor ’ s Solutions Manual Probability and Statistical
Inference Eighth Edition. Upper Saddle River, NJ(USA): Prentice Hall.
[45] Sincich, T., Stephan, D. & Levine, D. M. (2002). Practical statistics by example using Microsoft
Excel and Minitab. Prentice Hall Professional Technical Reference.
[46] Gopal, T. G. S. R. & Murali, K. (2015). A Critical Review on Factors Influencing Labour
Productivity in Construction. Journal of Mechanical and Civil Engineering (IOSR-JMCE),
12(5), 47–51.
[47] Heravi, G. & Eslamdoost, E. (2015). Applying Artificial Neural Networks for Measuring and
Predicting Construction-Labor Productivity. American Society of Civil Engineers, 141(10),
04015032.
[48] Duch, W. & Jankowski, N. (1999). Survey of neural transfer functions,” Neural Computing
Surveys, 2, 163–212.
[49] Nassar, Y. S. & Khaleel, T. A. (2019). Building Model to Predict Labour Productivity Using
Multiple Linear Regression Technique for " Formwork Concrete Columns ". Kufa Journal of
Engineering, 10(2), 12–26.

18