Machine learning techniques and multi-scale models to evaluate the impact of silicon dioxide (SiO2) and calcium oxide (CaO) in fly ash on the compressive strength of green concrete.pdf

Construction and Building Materials 400 (2023) 132604
Available online 29 July 2023
0950-0618/© 2023 Elsevier Ltd. All rights reserved.
Machine learning techniques and multi-scale models to evaluate the impact
of silicon dioxide (SiO2) and calcium oxide (CaO) in fly ash on the
compressive strength of green concrete
Dilshad Kakasor Ismael Jaf a,*
, Payam Ismael Abdulrahman b
, Ahmed Salih Mohammed c,d
,
Rawaz Kurda e
, Shaker M.A. Qaidi f
, Panagiotis G. Asteris g,*
a
Civil Engineering Department, College of Engineering, Salahaddin University-Erbil, Kurdistan, Iraq
b
Civil Engineering Department, College of Engineering, Tishk International University-Erbil, Erbil, Kurdistan Region, Iraq
c
Civil Engineering Department, College of Engineering, University of Sulaimani, Kurdistan, Iraq
d
The American University of Iraq, Sulaimani (AUIS), Sulaimani, Kurdistan Region. Iraq
e
Department of Highway and Bridge Engineering, Technical Engineering College, Erbil Polytechnic University, Erbil 44001, Iraq
f
Department of Civil Engineering, College of Engineering, University of Duhok, Duhok, Kurdistan-Region, Iraq
g
Computational Mechanics Laboratory, School of Pedagogical and Technological Education, Athens, Greece
A R T I C L E I N F O
Keywords:
Artificial neural network
Concrete
CaO (%)
SiO2 (%)
Fly Ash
Soft Computing
A B S T R A C T
Fly ash is a by-product almost found in coal power plants; it is available worldwide. According to the hazardous
impacts of cement on the environment, fly ash is known to be a suitable replacement for cement in concrete. A lot
of carbon dioxide (CO2) is released during cement manufacturing. The investigations estimate that about 8 – 10%
of the total CO2 emissions are maintained by cement production. Since fly ash has nearly the same chemical
compounds as cement, it can be utilized as a suitable alternative to cement in concrete (green concrete). The
current study analyzes the effect of the quantity of the two main components of fly ash, CaO, and SiO2, on the
compressive strength of concrete modified with different fly ash content for various mix proportions. For this
purpose, various concrete samples modified with fly ash were collected from the literature (236 datasets),
analyzed, and modeled using four different models; Full-quadratic (FQ), Nonlinear regression (NLR), Multi-linear
regression (MLR), and Artificial neural network (ANN) model to predict the compressive strength of concrete
with different geometry and size of the specimens. The accuracy of the models was evaluated using correlation
coefficient (R2
), Mean absolute error (MAE), Root mean squared error (RMSE), Scatter Index (SI), a-20 index, and
Objective function (OBJ). According to the modeling results, increasing SiO2 (%) increased the compressive
strength, while increasing CaO (%) increased compressive strength only when the cement replacement with fly
ash was between 52 and –100%. Based on R2
, RMSE, and MAE, the ANN model was the most effective and
accurate on predicting the compressive strength of concrete in different strength ranges. According to the
sensitivity analysis, curing time is the most critical characteristic for predicting the compression strength of
concrete using this database. The primary objective of this work is to explore and evaluate various machine-
learning models for predicting compressive resistance. The study emphasizes these models’ development,
comparison, and performance assessment, highlighting their potential to predict compressive strength accu
rately. The research primarily uses machine learning, leveraging algorithms and techniques to build predictive
models. The focus is on harnessing the power of data-driven approaches to improve the accuracy and reliability
of compressive strength predictions.
1. Introduction
Cement production provides 8 to 10 % of the CO2 emissions in the
world [1]. Cement is an important factor that leads to greenhouse gases
and global warming [2]. The key variables that lead to social and eco
nomic changes are industrialization, urbanization, globalization of the
* Corresponding authors.
E-mail addresses: dilshad.jaf@su.edu.krd (D. Kakasor Ismael Jaf), payam.ismael@alm.tiu.edu.iq (P. Ismael Abdulrahman), ahmed.mohammed@univsul.edu.iq
(A. Salih Mohammed), rawaz.kurda@epu.edu.iq (R. Kurda), shaker.abdal@uod.ac (S.M.A. Qaidi), asteris@aspete.gr (P.G. Asteris).
Contents lists available at ScienceDirect
Construction and Building Materials
journal homepage: www.elsevier.com/locate/conbuildmat
https://doi.org/10.1016/j.conbuildmat.2023.132604
Received 19 February 2023; Received in revised form 7 July 2023; Accepted 23 July 2023

2
economy, market and consumerism, and population increase. Overall,
climate change is a clear result of these factors resulting in dangerous
issues for humanity [3]. The rapid development of globalization and
population growth has caused an increase in building construction [4].
The construction business is growing exponentially, resulting in an
increasing demand for construction materials, especially cement [5].
Nowadays, the construction industry is one of the largest environmental
impacts among all human activities [6]. The negative environmental
impacts associated with the widespread use of cement have encouraged
researchers to find a satisfactory solution for reducing the demand for
cement. As the greatest strategy to reduce CO2 emissions is to replace
cement with a suitable substitute in percentage, several studies have
been done to find viable replacements for cement in concrete mixtures.
Supplementary cementitious materials (SCMs) like fly ash [7], silica
fume [8], ground granulated blast furnace slag [9], rice husk ash [10],
and metakaolin [11] can be added to the concrete as a partial replace
ment. Fly ash is a widely available material worldwide [12]. It is a very
fine powder with a fineness more than cement. Fly ash is commonly a
by-product material found in coal industries and collected in a baghouse
or on an electrostatic precipitator. Fly ash is one of the best alternatives
for cement in concrete; it provides excellent mechanical properties [13].
Utilizing fly ash in concrete economizes the mix, protects the environ
ment from disposal, conserves more land, and reduces cement produc
tion; overall, it can improve the properties of green and hardened
concrete [14–16].
Fly ash is a pozzolanic material rich in aluminous and siliceous
substances. Generally, the chemical compositions and their ratios are
source-dependent. According to the ASTM C619 [17], fly ash is divided
into two classes based on the quantity of chemical compositions; class C
and F [18,19]. Both classes of fly ash can be added to concrete in a
specific range to enhance mechanical properties and durability.
Numerous studies have been conducted on cement-based concrete
modified with different types, classes, and quantities of fly ash. The
improvement greatly depends upon the type and ratio of fly ash. The fly
ash type correspondence to the number of chemical substances and their
impact on the concrete properties. Among the chemical compositions of
fly ash, the effect of various silicon dioxide (SiO2) and calcium oxide
(CaO) ratios on the mechanical properties of concrete has been inves
tigated in previous studies. Class C of fly ash provides a greater
compressive strength value for the same cement dosage and fly ash
percentage than class F [19]. Fly ash with a higher CaO content tends to
have a higher reactivity [20]. The fly ash compounds such as CaO at
early ages and SiO2 at later ages are associated with the strength gain of
concrete [21].
Compressive strength is a significant mechanical property in con
crete considered in all concrete structure works. Concrete is a hetero
geneous material comprising different substances, such as cement,
aggregate, water, and admixtures, resulting in different compressive
strength values [22]. All the physical, chemical and mechanical char
acteristics of these ingredients and their quantity impact the compres
sive loading capacity of concrete [23]. The compressive strength value is
achieved in the laboratory during the testing of the concrete samples by
crushing the cylinders or cubes of standard dimension [24]. The
experimental method is standardized all over the world. However,
nowadays, laboratory tests are inefficient and uneconomical since it is
costly and time-consuming. Recently, with the advancement of Artificial
Intelligence (AI), Machine learning algorithm (ML), has been applied to
predict several mechanical properties in concrete [25]. ML techniques
such as regression, clustering, and classification can estimate various
factors with varying efficiency and accurately forecast concrete
compressive strength value [26]. Modeling the properties of concrete
can be developed in various ways, including statistical techniques,
computational modeling, and newly created tools like artificial neural
networks (ANN), regression analysis, and the M5P-tree model [27–31].
In the current study, 236 experimental values of compressive
strength have been collected from the literature. The concrete mixtures
included fly ash content (FA) ranging (from 71 – 316 kg/m3
), CaO (0.31
– 32%), SiO2 (30.5 – 62.54%), cement content (C) (67 – 356 kg/m3
),
cement replacement (CR) (18 – 100%), coarse aggregate (CA) (801 –
1246 kg/m3
), fine aggregate (S) (522 – 905 kg/m3
), water-to-binder
ratio (w/b) (0.28 – 0.6), and curing time (t) (3 – 365 days). As well as
the compressive strength (CS) ranged from 7.98 to 92.93 MPa. To check
the reliability and applicability of the models, CS is divided into three
different ranges; 5 to 34 MPa, 35 to 64 MPa, and 65 to 95 MPa. Then the
most effective model is applied to those ranges. This study proposes
models to predict compressive strength values using different analyzing
approaches. The influence of CaO and SiO2 contents at various mix
proportions on the compressive strength of cement-based concrete
modified with fly ash was studied and qualified in this study. In the
modeling part, full quadratic (FQ), nonlinear (NLR), multi-linear (MLR),
and artificial neural network (ANN) models are used to develop pre
dictive models. In addition, different statistical tools such as Correlation
coefficient (R2
), Root mean squared error (RMSE), Mean absolute error
(MAE), Scatter index (SI), Objective function (OBJ), and an a20-index
are used to evaluate the reliability and accuracy of the developed
models.
2. Research significant
The following are the general objectives of this study:
i. Statistically analyzing the collected data and evaluating the effect
of CaO and SiO2 on the compressive strength of fly ash-modified
concrete, such as coarse aggregate, fine aggregate, cement,
cement replacement, water-to-binder ratio (w/b), fly ash, cal
cium oxide (CaO,%), silicon dioxide (SiO2,%), and curing time.
ii. Creating an accurate and reliable model for predicting the
compressive strength of fly ash-based concrete.
iii. Conducting sensitivity analysis to determine the most important
parameter in predicting the compressive strength of concrete
modified with various types and classes of fly ash.
3. Methodology
The methodology of the current study includes a series of steps which
has been drawn as a flowchart and shown in Fig. 1. The main steps
involve the following;
i. Creating and collecting data on the cement-based concrete in
different strength ranges of (5 – 34 MPa), (35 – 64 MPa), and (65
– 95 MPa) from the literature.
ii. Considering coarse aggregate (CA), fine aggregate (S), cement
content (C), fly ash content (FA), cement replacement (CR),
water-to-binder ratio (w/b), calcium oxide (CaO), silicon dioxide
(SiO2), and curing time (t) as predictors for the models whereas,
the compressive strength (CS) is considered as a target value of
the models.
iii. Randomly mixing the collected data into two groups, 70% of the
data as training and 30% as testing.
iv. Developing predictive models using Full Quadratic (FQ),
Nonlinear Regression (NLR), Multi-linear Regression (MLR), and
Artificial Neural Networks (ANN) models.
v. Evaluating the proposed models based on R2
, RMSE, MAE, OBJ,
SI, and the a-20 index.
vi. Conducting sensitivity analysis to find the essential parameter in
predicting compressive strength for the concrete modified with
fly ash.
D. Kakasor Ismael Jaf et al.

3
3.1. Data collection
To develop a reliable model to predict the compressive strength of
concrete modified with fly ash, 236 data were collected from the liter
ature, including various fly ash ratios with class F and C, resulting in
different ratios of main chemical compositions such as; CaO and SiO2. As
well as different coarse and fine aggregate content, cement content, and
the curing regime was involved. Table 1 shows a summary of the
collected data. A total of 236 data were selected in this study, and based
on the information from the literature [32–34], the data were divided
into two groups using RAND Function.
Training data was included;0.70*No.ofdataset(236) = 165 Data.
The training data was used to develop the models.
Tested data; 0.30*No.ofdatatset(236)= 71 Data The testing dataset
was used to validate (Check) the developed equations based on 165
datasets.
3.2. Pre-processing
Pre-processing is required before using the dataset. Pre-processing
techniques improve the performance of computational models. As a
result, all discontinuous features (independent variables) were con
verted to dummy variables using Pandas, an open-source Python pack
age. As demonstrated in the box plot (Fig. 2), some features in the
created dataset had values that spanned an exceptionally wide range,
which was incompatible with model learning. Therefore, the average of
the upper and lower bounds was employed instead. Furthermore, using
the original non-normalized data may have influenced the ML model’s
performance [35]. This study pre-processed the database by converting
each independent variable to a value between zero and one using the
following equation, Eq. (1).
Xf =
(Xi − Ximin)
(
Xf max − Xf min
)
(Ximax − Ximin)
+ Xf min (1)
where; Xi is the old value and Xf is the new value. The Xf min is zero and
Xf max is one.
3.3. Statistical analysis
The collected data is statistically analyzed to determine the distri
bution of each independent variable with the dependent variable
(compressive strength). Therefore, statistical parameters such as Mean,
Standard Deviation (SD), Variance (Var), Kurtosis (Kur), Skewness
(Skew), minimum (Min), and Maximum (Max) are obtained. For kur
tosis, the negative value indicated a short distribution tail, while the
positive indicated a longer tail. However, skewness shows the distribu
tion of the variable, whether in the right or the left tail, with the right for
positive and the left for negative values.
The relation between compressive strength and the mentioned in
dependent variables is plotted in Fig. 3. The compressive strength data is
shown in a frequency histogram in Fig. 4. The statistical analysis is
summarized in Table 2.
Fig. 1. Methodology of Research; ANN, artificial neural network; FQ, full quadratic; NLR, nonlinear; MLR, multi-linear; R2
, correlation coefficient; RMSE, root mean
squared error; MAE, mean absolute error; SI, scatter index; OBJ, objectives.

4
3.4. Specimen sizes
The specimen parameters such as size and shape are important fac
tors in the compressive strength of concrete. Those factors were inves
tigated in the normal and high-strength concrete in the previous study
[36,37]. In the current study, the database utilized to create the models
comprised test samples of different sizes and shapes for assessing
compressive strength. Standard cylinders with S1 (100 × 200) and S2
(150 × 300) mm, and cubes with S3 (100 × 100 × 100) and S4 (150 ×
150 × 150) mm were included.
The samples were all utilized to create predictive models. Using
different models, the constructed models were evaluated using various
specimen sizes to determine the efficiency of predicted compressive
strength. 39% of the data were S1, and 18%, 41%, and 2% were S2, S3,
and S4, respectively. The total number for all specimen sizes in the
training and testing datasets is shown in Fig. 5.
Fig. 2. Box plot of the independent variables.
Table 1
Summary of collected data for cement-based concrete modified with fly ash.
Reference No.
of
Data
Coarse
aggregate,
CA (kg/m3
)
Sand, S
(kg/m3
)
Cement, C
(kg/m3
)
Fly ash
(kg/m3
)
Cement
replacement,
CR (%)
Water-to-
binder
ratio, (w/
b)
CaO (%) SiO2 (%) Curing
time, t
(Days)
Compressive
strength, CS
(MPa)
[58] 53 801–––1090 687–905 144–356 92–168 35–100 0.41–0.60 2.32 52.1 3, 7, 14,
28, 56 and
91
11.78–45.99
[15] 60 1270 522–671 160–320 80–240 20–60 0.30–0.40 1.54 62.54 7, 28, 56,
90,180,
256 and
365
16.99–92.93
[59] 42 1016–1207 607–854 107–328 71–316 18–74 0.3–0.4 27.6–31.9 30.5–34.9 3, 7, 28,
90 and
365
15.4–72.4
[60] 13 1089–1124 726–749 179–182 219–222 55 0.3–0.4 0.3–15.8 36.9–56.8 3, 7, 28,
56, and 91
21.2–44.6
[61] 9 843–864 842–866 161–247 155–254 40–60 0.4–0.5 13.4 52.4 7 and 28 15.6–48.3
[62] 23 1200 600 120–200 200–280 50–70 0.3–0.4 2.4–2.6 44.9–50.2 3, 7, 28,
90, 180,
and 365
7.98–81.54
[63] 15 1181–1246 600–819 66.9–200 167–280 50–80 0.28–0.40 2.6–26.3 35.2–50.2 7, 14, 28,
90, 180
and 365
16–67.5
[64] 17 1095–1195 740–797 150–274 117–204 30–52 0.4–0.5 4.2–13.4 42.7–52.4 7, 28, 56,
and 96
25.5–42.5
[65] 4 1114–1135 727–740 157–238 159–236 40–60 0.3 2.9 52 14 and 28 38–62
Remark 236 Ranged
between 801
and 1200
Ranged
between
522 and
905
Ranged
between
107 and
356
Varied
between
71 and
316
Ranged
between 18
and 100
Varied
between
0.28 and
0.6
Ranged
between
0.3 and
31.9
Ranged
between
30.5 and
62.54
Ranged
between 3
and 365
Ranged
between 7.98
and 92.93

5
Fig. 3. Marginal plots for the compressive strength of concrete with (a) coarse aggregate (kg/m3
) (b) fine aggregate (kg/m3
) (c) cement (kg/m3
) (d) fly ash (kg/m3
)
(e) cement replacement (%) (f) water-to-binder ratio (g) CaO (%) (h) SiO2 (%), and (i) curing time (days).

6
Fig. 3. (continued).

7

8

9
Fig. 4. Histogram for compressive strength (MPa) of concrete modified with fly ash.

10
3.5. Modeling
The correlation between dependent and independent variables es
tablishes a direct relationship between the concrete mixture proportion
and compressive strength. The datasets were randomly divided into
training and testing datasets [38,39]. The training dataset is used to
develop the models, whereas the testing dataset checks those developed
models against unobserved data during training. Based on the correla
tion matrix, a poor correlation between independent variables such as
CA, S, C, FA, CR, w/b, CaO, and SiO2 and the dependent variable which
is CS is observed as shown in Fig. 6. The correlations are 0.47, − 0.30,
0.40, − 0.23, − 0.53, − 0.58, − 0.08, and − 0.34, respectively. However,
an acceptable correlation coefficient between CS and t is noticed by
0.75. Therefore, different multivariable models are advanced in the
following section to establish an analytical model to predict the
compressive strength of concrete.
3.5.1. Full quadratic (FQ) model
Equation (2) shows the full quadratic formula, representing a rela
tionship between compressive strength, the first and second degrees of
each independent variable, and the interaction between independent
variables [32,40–44]. This model is found to have a complex combina
tion of mathematical terms [45,46].
CS = β1 + β2(CA) + β3(S) + β4(C) + β5(FA) + β6(CR) + β7
(w
b
)
+ β8(CaO) + β9(SiO2) + β10(t) + β11(CA)(S) + β12(CA)(C)
+ β13(CA)(FA) + β14(CA)(CR) + β15(CA)
(w
b
)
+ β16(CA)(CaO)
+ β17(CA)(SiO2) + β18(CA)(t) + β19(S)(C) + β20(S)(FA)
+ β21(S)(CR) + β22(S)
(w
b
)
+ β23(S)(CaO) + β24(S)(SiO2) + β25(S)(t)
+ β26(C)(FA) + β27(C)(CR) + β28(C)
(w
b
)
+ β29(C)(CaO)
+ β30(C)(SiO2) + β31(C)(t) + β32(FA)(CR) + β33(FA)
(w
b
)
+ β34(FA)(CaO) + β35(FA)(SiO2) + β36(FA)(t) + β37(CR)
(w
b
)
+ β38(CR)(CaO) + β39(CR)(SiO2) + β40(CR)(t) + β41
(w
b
)
(CaO)
+ β42
(w
b
)
(SiO2) + β43
(w
b
)
(t) + β44(CaO)(SiO2) + β45(CaO)(t)
+ β46(SiO2)(t)+β47(CA)2
+ β48(S)2
+ β49(C)2
+β50(FA)2
+β51(CR)2
+ β52
(w
b
)2
+ β53(CaO)2
+β54(SiO2)2
+ β55(t)2
(2)
CS, CA, S, C, FA, CR, w/b, CaO, SiO2, and t are compressive strength,
coarse aggregate content, fine aggregate content, cement content, fly
ash content, cement replacement, water-to-binder ratio, calcium oxide
Table 2
Summary of statistical analysis of the concrete parameters.
Variables Statistical parameters
Mean Standard deviation Variance Kurtosis Skewness Min Max
CA (kg/m3
) 1132.2 120.73 14575.73 − 0.265 − 0.666 801 1270
S (kg/m3
) 709.1 100.57 10114.55 − 1.2776 − 0.0785 522 905.4
C (kg/m3
) 211.2 59.809 3577.099 − 0.432 0.240 67 356
FA (kg/m3
) 173.7 58.262 3394.479 − 0.827 0.278 71 316
CR (%) 49.9 15.506 240.440 0.264 0.119 18 100
w/b 0.39 0.083 0.007 − 0.324 0.732 0.28 0.60
CaO (%) 8.75 10.929 119.444 − 0.252 1.232 0.31 31.9
SiO2 (%) 49.42 10.755 115.676 − 0.890 − 0.463 30.5 62.54
t (Dyas) 86.9407 110.848 12287.359 1.071 1.520 3 365
CS (MPa) 40.75 18.52 342.98 0.0174 0.77 7.98 92.93
Fig. 5. A summary of the sample sizes utilized in training and testing for developing models.

11
ratio, silicon dioxide rato, and curing time, respectively. Furthermore,
the β1 to β55 are model parameters.
3.5.2. Nonlinear regression (NLR) model
The following formula is a typical form to develop a nonlinear
regression model [32,47,48] for predicting the compressive strength of
concrete containing different types of fly ash with various quantities of
CaO and SiO2, Eq. (3). NLR model has a disadvantage which is the
complex mathematical operation [49].
CS, CA, S, C, FA, CR, w/b, CaO, SiO2, and t are compressive strength,
coarse aggregate content, fine aggregate content, cement content, fly
ash content, cement replacement, water-to-binder ratio, calcium oxide
ratio, silicon dioxide ratio, and curing time, respectively. The α1 to α18 is
defined as the model parameters.
3.5.3. Multi-linear regression (MLR) model
The compressive strength of the concrete containing different CaO
and SiO2 ratios is predicted using the multi-linear regression model.
Equation (4) represents the product of the variables affecting the
compressive strength of the concrete in exponential and constant terms.
The MLR model has some advantages, such as simple mathematical
operation and easy implementation resulting in more accurate results
compared to the LR model. However, the MLR model greatly depends on
the number of data; fewer number provides less accuracy. Therefore,
this model has poor quality [32].
CS = α1(CA)α2
(S)α3
(C)α4
(FA)α5
(CR)α6
(w
b
)α7
(CaO)α8
(SiO2)α9
(t)α10
(4)
Where α1toα10 are defined as the model parameters
βn = an(CA) + bn(S) + cn(C) + dn(FA) + en(CR) + fn
(w
b
)
+ gn(CaO)
+ hn(SiO2) + in(t) + b
(5a)
CS =
Node1
1 + e− β1
+
Node2
1 + e− β2
+ ⋯ +
Noden
1 + e− βn
+ Threshold (5b)
3.5.4. Artificial neural network (ANN model)
The Artificial Neural Network (ANN) is a computational system that
analyzes data like the human brain. This model is also a machine
learning system used in construction engineering for various numerical
forecasts and challenges [27,29,31,42]. ANN is made up of three layers;
input, hidden, and output, which are linked by biases and weights
[44,50], Eq. (5). This study designed a multi-layer feed-forward network
with concrete components (CA, S, C, FA, CR, w/b, CaO, SiO2, and t) as
input and CS as output. In the output layer, a sigmoid activation function
is utilized.
Output = f
(
∑
n
j=1
wjxj + bias
)
(5)
j is the number of input variables, xj is the input parameter, and bias is
the threshold for the sigmoid activation function.
Fig. 7 shows the typical process of the ANN result. As well as the
typical procedure for ANN calculation in a single hidden layer is illus
trated in Eq. (5).
The Node1 to Noden, and Threshold are directly calculated.
3.6. Metrics to evaluate developed models
The developed models are evaluated using different parameters such
as correlation coefficient (R2
), root mean squared error (RMSE), mean
absolute error (MAE), scatter index (SI), objectives (OBJ), and a20-
index. The values of the a20-index are predicted to be unity for a per
fect predictive model. The proposed a20-index has the advantage of
having physical engineering meaning; it declares the number of samples
that satisfy expected values with a 20% variance from experimental
values [51]. The following equations (6) – (11) are utilized to determine
each mentioned criterion [46,52–54].
CA
S
C
FA
CR
w/b
CaO
t
CS
CA S C FA CR w/b CaO t CS
SiO2
SiO2
Fig. 6. Correlation matrix plot between the dependent and independent variables of concrete modified with fly ash.
CS = α1(CA)α2
+ α3(S)α4
+ α5(C)α6
+α7(FA)α8
+α9(CR)α10
+ α11
(w
b
)α12
+ α13(CaO)α14
+α15(SiO2)α16
+ α17(t)α18
(3)

12
R2
=
⎛
⎜
⎜
⎝
∑
i(yp − x) × (yi − y)
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑
i
(yp − x)2
√
×
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑
i
(yi − y)2
√
⎞
⎟
⎟
⎠
2
(6)
RMSE =
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑
n
i=1
(yi − yp)2
N
√
√
√
√
√
(7)
MAE =
∑n
i=1(|yi − yp|)
n
(8)
SI =
RMSE
yi
(9)
OBJ =
(
ntr
nall
×
RMSEtr + MAEtr
R2
tr + 1
)
+
(
ntst
nall
×
RMSEtst + MAEtst
R2
tst + 1
)
(10)
a20 − index =
N20
N
(11)
where;
yp = predicted value of compressive strength.
yi = experimental value of compressive strength.
x = mean of predicted compressive strength values.
y = mean of experimental compressive strength data.
ntr = number of the training dataset.
ntst = number of the testing dataset.
nall = a total number of datasets, including training and testing.
N = total number of data in the dataset.
N20 = total number of predicted to the experimental data of
compressive strength ratio ranged from 0.8 to 1.2.
The value of R2
and a-20 index is commonly between (0to1), and the
best value is considered 1. The RMSE, MAE, and OBJ values are between
(0to∞), the value is recommended to be low as possible, and zero is the
best value. As well as, if the value of SI was smaller than 0.1, the model is
qualified as an excellent performance. Whereas the SI value is between
(0.1 to 0.2), (0.2 to 0.3), and greater than 0.3, indicating the good, fair,
and poor performance of the model, respectively [44,55].
4. Results and discussion
4.1. Variation of predicted and measured compressive strength
4.1.1. FQ model
The full quadratic model is one of the most effective models due to its
advanced mathematical expression. This model was used to predict the
compressive strength of concrete modified with various fly ash types. It
was derived based on mathematical parameters such as constant, linear,
variable product terms/interaction, and squared variables. Equation
(12) shows the formula for the FQ model for predicting the compressive
strength of concrete. The correlation between predicted and measured
compressive strength for the FQ model is illustrated in Fig. 8. The value
of R2
and RMSE was 0.94 and 4.627 for training and 0.887 and 4.57 for
testing data, respectively. The training dataset contains an error line of
15 to − 15%, meaning that 85% of the data falls between 0.85 and 1.15
for the predicted to measured compressive strength ratio.
CS = − 1.26 − 0.17(CA) − 0.03(S) + 0.9(C) − 0.35(FA) − 1.36(CR)
+ 1.2
(w
b
)
− 2.8(CaO) + 5.46(SiO2) + 0.4(t) + 0.0002(CA)(S)
− 0.0002(CA)(C) − 0.0002(CA)(FA) + 0.001(CA)(CR)
− 0.05(CA)
(w
b
)
+ 0.0001(CA)(CaO) − 0.003(CA)(SiO2)
− 0.0003(CA)(t) − 0.0004(S)(C) − 0.001(S)(FA) + 0.00003(S)(CR)
− 0.19(S)
(w
b
)
+ 0.003(S)(CaO) + 0.0012(S)(SiO2) − 0.0004(S)(t)
+ 0.002(C)(FA) − 0.0003(C)(CR) − 0.18(C)
(w
b
)
− 0.0007(C)(CaO)
− 0.0097(C)(SiO2) − 0.0002(C)(t) + 0.004(FA)(CR) + 0.48(FA)
(w
b
)
− 0.004(FA)(CaO) − 0.005(FA)(SiO2) − 0.001(FA)(t)
− 0.16(CR)
(w
b
)
+ 0.04(CR)(CaO) + 0.002(CR)(SiO2)
+ 0.003(CR)(t) + 1.95
(w
b
)
(CaO) + 0.04
(w
b
)
(SiO2) − 0.26
(w
b
)
(t)
− 0.03(CaO)(SiO2) + 0.001(CaO)(t) + 0.6(SiO2)(t)
+ 0.0001(CA)2
− 0.000002(S)2
− 0.0005(C)2
+ 0.001(FA)2
− 0.0003(CR)2
+ 0.0001
(w
b
)2
+ 0.0001(CaO)2
+ 0.0004(SiO2)2
− 0.0003(t)2
(12)
Fig. 7. Typical ANN output technique in a single node.

13
No. of training dataset = 165, R2
= 0.94, RMSE = 4.627 MPa.
4.1.2. NLR model
The nonlinear regression model was also used to predict the
compressive strength of fly ash-modified concrete. The NLR model result
is shown in Eq. (13). The relationship between predicted and measured
compressive strength is displayed in Fig. 9. The training dataset has an
R2
and RMSE value of 0.816 and 8.36 MPa. Also, an R2
of 0.47 and RMSE
of 9.62 MPa for the testing dataset were observed. According to the
RMSE results, the NLR model performance is fair. The FQ model results
provide better performance than the NLR model. The error line in the
training dataset is 30 to − 30%, which means that 70% of the data falls
between 0.7 and 1.3 for the predicted to measured compressive strength
ratio.
= 0.816, RMSE = 8.36 MPa.
4.1.3. MLR model
The multi-linear regression model is another model utilized to pre
dict the compressive strength of concrete modified with fly ash. It is
counted as one of the least effective models due to its elementary
mathematical expression. The MLR model formula consists of constant
terms with terms that rose to the power of constant variables. The
following equation presents the variables and their relationship in
detail; Eq. (14). The relation between predicted and measured
compressive strength is illustrated in Fig. 10. The R2
and RMSE were
0.79 and 9.22 for training, 0.61 and 8.25 for testing data, respectively.
The error line in the training dataset is 25 to − 25%, which means that
75% of the data falls between 0.75 and 1.25 for the projected to
measured compressive strength ratio.
= 0.79, RMSE = 9.22 MPa.
4.1.4. ANN model
The Optimum ANN network structure is the last model used to pre
dict the compressive strength of concrete. Five trials were selected
depending on the RMSE and MAE values, as shown in Fig. 11. Among
those five trials, the lowest RMSE and MAE values were observed in the
CS = 140.66 (CA)− 0.008
+ 0.68(S)− 2.47
+ 1.06(C)0.0002
− 0.041(FA)0.002
− 3.096(CR)− 6.87
+ 13.94
(w
b
)− 1.14
− 26.95 (CaO)− 0.004
− 227(SiO2)− 0.146
+ 5.32 (t)0.35
(13)
Fig. 8. Comparison between measured and predicted CS using FQ model for training and testing dataset.
CS= 1.3 (CA)0.8
(S)0.35
(C)0.16
(FA)0.18
(CR)− 0.74
(w
b
)− 0.05
(CaO)0.37
(SiO2)0.2
+ (t)0.0000002
(14)

14
network containing one hidden layer with 19 neurons. The ANN
network having one hidden layer with 19 neurons with 0.1 of mo
mentum, a learning rate of 0.2, and 2000 learning time, is chosen,
Fig. 12. The ANN formula, including weights and biases, is shown in Eq.
(15). Fig. 13 shows the relation between predicted compressive strength
by the ANN having 19 neurons with the experimental compressive
strength values for both training and testing data. ANN network analysis
has an R2
of 0.987 and RMSE of 3.3 MPa when the training data are
applied. The model provides an R2
of 0.986 and RMSE of 3.52 MPa when
testing data is used. The training dataset has an error line of 10 to − 10%,
indicating that 90% of the data are laid between 0.9 and 1.1 for the ratio
of predicted compressive strength to measured compressive strength. As
well as Fig. 13 shows the variation between measured and predicted
compressive strength values plotted for different strength ranges based
on the training dataset. The highest value of R2
was found in the CS of 65
to 95 MPa.
CS = −
0.1
1 + e− β1
−
0.02
1 + e− β2
−
0.15
1 + e− β3
− ⋯
0.95
1 + e− β19
+ 0.475 (15)
= 0.987, RMSE = 3.3 MPa.
4.2. Effect of CaO and SiO2 content on the compressive strength of
concrete
The FQ model was used to examine the influence of cement
replacement percentage, CaO (%), and SiO2 (%) on cement-based con
crete’s compressive strength using a fly-ash modified concrete dataset.
All independent parameters were held constant, and the CR was
increased from 18 to 100%; the compressive strength grew as the cement
replacement increased. The SiO2 (%) was then increased from 30.5 to
62.54% while the CA, S, C, FA, CR, w/b, CaO, and t remained constant.
According to the model results, increasing SiO2 causes a gradual in
crease in compressive strength, while CaO (%) ranges between 0.3 and
0
20
40
60
80
100
0 20 40 60 80 100
Training Testing
Compressive Strength,σc Measured (MPa)
Compresive
Strength,
σc
Predicted
(MPa)
NLR Model (Eq.13)
Training Data = 165
R2 = 0.816
RMSE = 8.36 MPa
-30%
Error
+
3
0
%
E
r
r
o
r
Y=X
line
NLR Model (Eq.13)
Testing Data = 71
R2 = 0.47
RMSE = 9.62 MPa
Fig. 9. Comparison between measured and predicted CS using NLR model for training and testing dataset.
− 0.699 − 0.034 0.187 0.099 0.278 0.126 0.156 − 0.032 0.423 − 1.770 β1
− 0.674 − 0.024 0.141 0.107 0.324 − 0.007 0.186 0.074 0.431 − 1.815 β2
− 0.719 − 0.082 0.065 0.552 0.144 0.343 0.227 − 0.089 0.294 − 1.644 β3
− 0.9 0.062 0.136 − 0.093 0.332 0.132 − 0.098 − 0.26 0.482 − 1.673 β4
− 2.249 − 0.093 0.085 0.059 − 0.422 − 0.77 − 1.588 1.53 0.384 − 2.852 β5
− 0.787 0.044 0.099 0.052 0.353 0.371 − 0.033 − 0.262 0.355 − 1.656 CA β6
− 0.618 0.074 − 0.137 0.399 0.459 0.824 − 0.182 − 0.597 0.042 − 1.569 S β7
− 0.73 − 0.084 0.111 0.382 0.201 0.36 0.248 − 0.048 0.415 − 1.618 C β8
1.067 2.077 0.468 − 1.587 − 0.042 − 2.883 − 0.566 − 0.952 0.889 − 3.131 FA β9
− 0.406 − 0.46 − 0.089 0.705 − 0.152 − 0.003 1.294 − 0.354 − 2.561 − 1.826 × CR = β10
− 0.152 − 0.16 1.543 0.415 − 1.948 1.465 − 1.387 0.046 − 1.439 − 2.373 w/b β11
− 1.143 0.245 0.73 − 0.905 0.052 − 0.04 − 0.104 0.063 1.198 − 1.805 CaO β12
− 0.738 − 0.034 0.023 0.399 0.242 0.285 0.069 − 0.22 0.281 − 1.661 SiO2 β13
− 0.668 − 0.034 0.176 0.143 0.353 − 0.051 0.166 0.082 0.432 − 1.823 t β14
− 0.78 − 0.06 0.129 0.086 0.294 0.113 0.087 − 0.125 0.394 − 1.743 1 β15
0.151 0.712 − 1.587 2.137 0.13 1.045 − 1.958 1.349 0.004 − 3.212 β16
− 0.698 − 0.092 0.07 0.326 0.213 0.276 0.283 0.036 0.479 − 1.556 β17
− 0.714 − 0.083 0.061 0.538 0.143 0.278 0.338 0.059 0.446 − 1.637 β18
− 0.734 − 0.071 0.145 0.186 0.0.274 0.293 0.151 − 0.129 0.373 − 1.678 β19

15
Fig. 10. Comparison between measured and predicted CS using MLR model for training and testing dataset.
Fig. 11. Selecting optimum ANN based on RMSE and MAE.

16
Fig. 12. Optimal ANN network structures containing one hidden layer and 19 hidden neurons.

17
Fig. 13. Comparison between measured and predicted CS using ANN model for training and testing dataset.

18
24% but contributes a reduction above that. To determine the influence
of CaO (%), all other parameters were fixed, including CA, S, C, FA, CR,
w/b, and SiO2, and the CaO (%) varied from 0.3 to 31.9%. Concrete
compressive strength improved as the CaO (%) increased, but only when
the cement replacement was between 52 and 100% and dropped below
that. Fig. 14 depicts the performance of CaO and SiO2.
5. Evaluation of developed models
The study attempted to determine the effect of two major chemical
compositions of fly ash, CaO, and SiO2, on the compressive strength of
cement-based concrete modified with various types and amounts of fly
ash. The experiment includes forecasting concrete compressive strength
using four alternative models; FQ, NLR, MLR, and ANN. Each model
supplied a formula based on several mathematical parameters. To
evaluate the performance of each constructed model, various assessment
criteria were applied.
According to the R2
, RMSE, and MAE statistics, the ANN model has the
maximum accuracy and reliability for prediction using the training and
testing dataset. Table 3 summarizes the statistical criteria outcomes for
the created models. The ANN model has an R2
of 0.987, RMSE of 3.3 MPa,
and MAE of 2.662 MPa based on the training dataset, as well as R2
of
0.986, RMSE of 3.52 MPa, and MAE of 2.96 MPa for the testing dataset.
Furthermore, more data are along the Y = X line for the ANN model,
which has an error line of 10 to − 10% for the training dataset, referring
to the fact that 90% of the data are between 0.90 and 1.1 (predicted CS/
measured CS). The second-ranked model is the FQ model; it has an R2
of
0.94, RMSE of 4.627 MPa, and MAE of 3.496 MPa for the training
dataset, and R2
of 0.887, RMSE of 4.57 MPa, and MAE of 3.457 for the
testing dataset. The FQ model has an error line of ± 15 for the training
dataset. The NLR and MLR model have error lines of ± 30, and ± 25,
respectively, indicating the model’s poor performance. Fig. 15 compares
created models based on the testing dataset; model values are within the
± 20 and ± 30% error lines.
The training dataset for the models based on three different CS
ranges was plotted, and the RMSE values were provided. Table 4 shows
an overview of the findings. There were 63 datasets for the strength
range of 5 to 34 MPa, 78 in the range of 65 to 95 MPa, and 24 data for the
65 to 95 MPa. According to the findings, the ANN model was more
appropriate than other models for all CS ranges. The model provided an
RMSE of 3.576, 3.16, and 2.616 MPa for 5 to 34, 35 to 65, and 65 to 95
MPa, respectively.
The same models were employed to identify the ideal specimen
shape and size. The dataset for each size specimen was modeled based
on training and testing datasets. Table 5 shows RMSE values for different
models. According to the table, based on a satisfied number of data, the
cylinder specimens with dimensions of 100*200 mm have the lowest
RMSE value, 0.124 MPa, for the training dataset using the ANN model.
Although, the cylinder specimens with 150*300 mm dimensions have
the lowest value of RMSE for the testing dataset, which is 0.01 MPa.
Furthermore, the created models using the training and testing
dataset were also evaluated by comparing the OBJ function and SI
values. The highest value of SI is 0.104 for training, and the ANN model,
Fig. 16, concluded 0.074 for testing. The model also has the lowest OBJ
function compared to the other models, indicating that the model is
more accurate in forecasting the compressive strength of concrete
modified with fly ash. The model has an OBJ value of 3.09 and 1.0 MPa
based on the training and testing dataset, respectively, Fig. 17.
Although, the highest a-20 index value was observed for the MLR model;
104.4% for the training and 104.8% for the testing dataset, Fig. 18.
Fig. 14. The effect of CaO and SiO2 content of fly ash on the compressive
strength of concrete.
Table 3
Summary of performance evaluation parameters for the developed models.
Models Fig. (No) Eq. (No.) Training Testing Ranking
R2
RMSE (MPa) MAE (MPa) R2
RMSE (MPa) MAE (MPa)
FQ 8 12 0.94 4.63 3.496 0.887 4.57 3.457 2
NLR 9 13 0.816 8.36 6.698 0.47 9.62 7.522 3
MLR 10 14 0.79 9.22 7.199 0.62 8.25 6.665 4
ANN 13 15 0.987 3.3 2.662 0.986 3.52 2.96 1
Fig. 15. Variation of measured and predicted CS for developed models using
the testing dataset.
Table 4
Root mean squared error (RMSE) for the developed models in different ranges of
compressive strength.
Compressive
Strength range
(MPa)
No.
of
data
Models Best Model
Performance
FQ NLR MLR ANN
5 to 34 63 3.731 7.723 6.335 3.576 ANN
35 to 64 78 4.312 8.231 10.599 3.16 ANN
65 to 95 24 7.087 10.21 10.686 2.616 ANN

19
Fig. 16. Comparison of developed models using (a) SI and (b) MAE.
Table 5
Root mean squared error (RMSE) of the developed models using different specimen sizes and conducting.
Dataset Sample size (mm) No. of Samples Models Best Model Performance
FQ NLR MLR ANN
Training S1(100*200) 58 3.746 4.757 5.479 1.124 ANN
S2(150*300) 30 2.363 6.592 6.008 2.512 ANN
S3(100*100*100) 74 3.663 8.48 5.428 1.94 ANN
S4(150*150*150) 3 NA 3.673 11.2 0 ANN
Testing S1(100*200) 34 3.864 6.696 5.287 1.116 ANN
S2(150*300) 12 10.179 6.592 5.196 0.01 ANN
S3(100*100*100) 24 6.108 12.186 8.377 1.94 ANN
S4(150*150*150) 1 N.A 6 3.54 0 ANN

20
In addition, the created models were compared using residual error,
as shown in Fig. 19. The residual error value is obtained by subtracting
the expected compressive strength from the measured compressive
strength. The results show that the FQ model has the lowest error value,
ranging from − 16.58 to + 17.14 MPa. At the same time, the residual
error for the NLR, MLR, and ANN models is − 17.81 to + 24.64 MPa,
− 24.61 to + 24.85 MPa, and − 6.80 to + 8.28 MPa, respectively.
6. Sensitivity investigation
A sensitivity analysis was performed to evaluate the influence of each
independent parameter on the value of modeled dependent variable
[32,44,56,57]. To determine the sensitivity index of each parameter on
the developed model, one parameter was extracted each time for the
training data. The ANN was used for the sensitivity investigation as it
was observed as the best model for predicting the compressive strength
of concrete compared to the other conducted models. All R2
, RMSE, and
MAE values were reported for each trial. The sensitivity results are
shown in Table 6. The findings indicate that the curing time is the most
significant and influential parameter for predicting the compressive
strength of concrete, Fig. 20.
7. Conclusions
To find an accurate and reliable model to predict the compressive
strength of cement concrete modified with different fly ash types and
quantities, 236 data samples for fly ash-modified cement concrete with
different mixture proportions, CaO and SiO2 quantity of fly ash, aggre
gate content, cement content, cement replacement, water-to-binder
ratio, and curing time were collected from previous researches. Ac
cording to the collected data and the result of four different model ap
proaches, the following can be drawn:
The cement replacement percentage ranged between 18 and 100%.
The compressive strength of concrete gradually increased when the
replacement (%) was increased.
The data collected from various research studies show that CaO (%)
varied between 0.3 and 31.9%. SiO2 (%) ranged between 30.5 and
62.54%.
Increasing CaO (%) in fly ash caused an increase in compressive
strength only when the cement replacement ranged between 52 and
100%. The CS also improved with increasing SiO2 (%), where; the
CaO (%) varied between 0.3 and 24%.
Based on multiple assessment criteria such as R2
, RMSE, and MAE,
the ANN model showed the best accuracy and performance for pre
dicting the compressive strength of concrete. The model had the
highest value of R2,
which was 0.987 and 0.986 for training and
testing datasets, respectively. The lowest value of RMSE and MAE
was found as 3.3 and 2.66 for training, and 3.53 and 2.96 for testing,
respectively.
Related to other statistical tools, such as SI and OBJ, the ANN model
was first ranked. The SI and OBJ function values were 0.104 and
3.09 MPa for training and 0.074 and 1.0 MPa for the testing dataset,
respectively. Although the highest a-20 index was observed for the
MLR model, the value was 104.4% for the training dataset.
According to the shape and size of specimens, the cylinder specimens
having 100*200 mm were obtained for the best performance using
the ANN model.
Fig. 17. Comparison of developed models using the OBJ function.
99.7%
100.4%
104.4%
97.2%
101.8%
96.9%
104.8%
91.4%
0%
20%
40%
60%
80%
100%
120%
FQ NLR MLR ANN
Training Testing
Models
a-20
index
(%)
(b)
Fig. 18. Comparison of developed models using a20-index.

21
Fig. 19. Variation between measured and predicted CS for developed models based on residual error (a) ANN and FQ, (b) ANN and NLR, and (c) ANN and MLR.

22
Table 6
Sensitivity analysis using the ANN model applied to the training dataset.
No. Combination Removed Parameter R2
RMSE (MPa) MAE (MPa) Ranking based on RMSE and MAE
1 CA, S, C, FA, CR, w/b, CaO, SiO2, t – 0.987 3.30 2.662 –
2 S, C, FA, CR, w/b, CaO, SiO2, t CA 0.987 3.984 3.254 3
3 CA, C, FA, CR, w/b, CaO, SiO2, t S 0.989 3.045 2.497 7
4 CA, S, FA, CR, w/b, CaO, SiO2, t C 0.990 3.800 3.165 4
5 CA, S, C, CR, w/b, CaO, SiO2, t FA 0.989 3.127 2.514 6
6 CA, S, C, FA, w/b, CaO, SiO2, t CR 0.990 3.800 3.165 4
7 CA, S, C, FA, CR, CaO, SiO2, t w/b 0.976 4.424 3.446 2
8 CA, S, C, FA, CR, w/b, SiO2, t CaO 0.990 2.800 2.217 8
9 CA, S, C, FA, CR, w/b, CaO, t SiO2 0.989 3.523 2.824 5
10 CA, S, C, FA, CR, w/b, CaO, SiO2 t 0.876 10.450 8.109 1
Fig. 20. The percentage contribution of input variables in predicting compressive strength of concrete using the ANN model.

23
The ANN model was applied to different CS ranges of concrete. The
best performance of the model was observed in the range of 35 to 64
MPa.
Sensitivity analysis illustrates the most effective parameter for pre
dicting the compressive strength of concrete: the curing time.
Using multiple models provides construction professionals with
broader information and insights. This enhances decision-making
processes by considering different perspectives and approaches,
leading to more informed and effective decisions throughout the
project lifecycle.
Employing multiple models allows for the validation and cross-
verification of results. By comparing the predictions and outcomes
of different models, construction practitioners can assess the accu
racy and reliability of the models in different scenarios, increasing
confidence in the results.
The construction industry is diverse, and no single model can address
all situations. Applying multiple models acknowledges this diversity
and provides the flexibility to choose the most appropriate model for
specific projects, materials, or contexts. It enables the construction
industry to adapt to evolving needs and challenges.
The use of multiple models encourages continuous improvement in
construction practices. By comparing and evaluating different
models, areas for improvement can be identified, leading to ad
vancements in modeling techniques, data collection, and analysis
methodologies. This fosters innovation and progress within the
industry.
Discuss the advantages of machine learning models in the context of
compressive strength prediction. These may include their ability to
capture complex relationships, handle large datasets, adapt to
different scenarios, and potentially provide better accuracy than
traditional analytical models.
Funding
This work had no finding.
CRediT authorship contribution statement
Dilshad Kakasor Ismael Jaf: . Payam Ismael Abdulrahman: .
Ahmed Salih Mohammed: . Rawaz Kurda: Investigation, Visualiza
tion. Shaker M.A. Qaidi: Validation. Panagiotis G. Asteris: Supervi
sion, Writing – original draft, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Data availability
Data will be made available on request.
References
[1] B. Suhendro, Toward green concrete for better sustainable environment, Procedia
Eng. 95 (2014) 305–320.
[2] M.E. Bildirici, Cement production, environmental pollution, and economic growth:
evidence from China and USA, Clean Techn. Environ. Policy 21 (4) (2019)
783–793.
[3] S. Joel, Compressive strength of concrete using fly ash and rice husk ash: A review,
Civ Eng J 6 (7) (2020) 1400–1410.
[4] Ganesh Prabhu, G., Bang, J. W., Lee, B. J., Hyun, J. H., & Kim, Y. Y. (2015).
Mechanical and durability properties of concrete made with used foundry sand as
fine aggregate. Advances in Materials Science and Engineering, 2015.
[5] T. Ayub, S.U. Khan, F.A. Memon, Mechanical characteristics of hardened concrete
with different mineral admixtures: a review, Scientific World Journal 2014 (2014)
1–15.
[6] I. Ignjatović, Z. Sas, J. Dragaš, J. Somlai, T. Kovács, Radiological and material
characterization of high volume fly ash concrete, J. Environ. Radioact. 168 (2017)
38–45.
[7] C. Herath, C. Gunasekara, D.W. Law, S. Setunge, Performance of high volume fly
ash concrete incorporating additives: A systematic literature review, Constr. Build.
Mater. 258 (2020), 120606.
[8] R. Siddique, Utilization of silica fume in concrete: Review of hardened properties,
Resour. Conserv. Recycl. 55 (11) (2011) 923–932.
[9] S.C. Pal, A. Mukherjee, S.R. Pathak, Investigation of hydraulic activity of ground
granulated blast furnace slag in concrete, Cem. Concr. Res. 33 (9) (2003)
1481–1486.
[10] R. Khan, A. Jabbar, I. Ahmad, W. Khan, A.N. Khan, J. Mirza, Reduction in
environmental problems using rice-husk ash in concrete, Constr. Build. Mater. 30
(2012) 360–365.
[11] P.-M. Zhan, Z.-H. He, Z.-M. Ma, C.-F. Liang, X.-X. Zhang, A.A. Abreham, J.-Y. Shi,
Utilization of nano-metakaolin in concrete: A review, Journal of Building
Engineering 30 (2020), 101259.
[12] V.M. Malhotra, High-performance high-volume fly ash concrete, Concr. Int. 24 (7)
(2002) 30–34.
[13] M. Amran, R. Fediuk, G. Murali, S. Avudaiappan, T. Ozbakkaloglu, N. Vatin,
M. Karelina, S. Klyuev, A. Gholampour, Fly ash-based eco-efficient concretes: A
comprehensive review of the short-term properties, Materials 14 (15) (2021) 4264.
[14] C.D. Atiş, High volume fly ash abrasion resistant concrete, J. Mater. Civ. Eng. 14
(3) (2002) 274–277.
[15] B. Kumar, G.K. Tike, P.K. Nanda, Evaluation of Properties of High-Volume Fly-Ash
Concrete for Pavements, Journal of Materials in Civil Engineering 19 (10) (2007)
906–911.
[16] S. Izquierdo, J. Diaz, R. Mejia, J. Torres, Blended cement containing fluid catalytic
cracking catalyst residue (FCC): hydration and paste microstructure, Revista
Ingeniería de Construcción 28 (2) (2013) 141–154.
[17] ASTM C618–19, Standard Specification for Coal Fly Ash and Raw or Calcined
Natural Pozzolan for Use in Concrete, ASTM International, West Conshohocken,
PA, 2019.
[18] R. Siddique, Performance characteristics of high-volume Class F fly ash concrete,
Cem. Concr. Res. 34 (3) (2004) 487–493.
[19] M. Sumer, Compressive strength and sulfate resistance properties of concretes
containing Class F and Class C fly ashes, Constr. Build. Mater. 34 (2012) 531–536.
[20] S. Aydın, H. Yazıcı, H. Yiğiter, B. Baradan, Sulfuric acid resistance of high-volume
fly ash concrete, Build. Environ. 42 (2) (2007) 717–721.
[21] T. Hemalatha, A. Ramaswamy, A review on fly ash characteristics–Towards
promoting high volume utilization in developing sustainable concrete, J. Clean.
Prod. 147 (2017) 546–559.
[22] W. Sarwar, K. Ghafor, A. Mohammed, Regression analysis and Vipulanandan
model to quantify the effect of polymers on the plastic and hardened properties
with the tensile bonding strength of the cement mortar, Results Mater. 1 (2019)
100011.
[23] D.M. Cotsovos, M.N. Pavlović, Numerical investigation of concrete subjected to
compressive impact loading. Part 2: parametric investigation of factors affecting
behaviour at high loading rates, Comput. Struct. 86 (1–2) (2008) 164–180.
[24] M. Li, H. Hao, Y. Shi, Y. Hao, Specimen shape and size effects on the concrete
compressive strength under static and dynamic tests, Constr. Build. Mater. 161
(2018) 84–93.
[25] D.-C. Feng, Z.-T. Liu, X.-D. Wang, Y. Chen, J.-Q. Chang, D.-F. Wei, Z.-M. Jiang,
Machine learning-based compressive strength prediction for concrete: an adaptive
boosting approach, Constr. Build. Mater. 230 (2020), 117000.
[26] A. Ahmad, F. Farooq, K.A. Ostrowski, K. Śliwa-Wieczorek, S. Czarnecki,
Application of novel machine learning techniques for predicting the surface
chloride concentration in concrete containing waste material, Materials 14 (9)
(2021) 2297.
[27] P. Sihag, P. Jain, M. Kumar, Modelling of impact of water quality on recharging
rate of storm water filter system using various kernel function based regression,
Modeling Earth Systems and Environment 4 (1) (2018) 61–68.
[28] W. Mahmood, A. Mohammed, New Vipulanandan pq model for particle size
distribution and groutability limits for sandy soils, J. Test. Eval. 48 (5) (2019)
3695–3712.
[29] W. Qadir, K. Ghafor, A. Mohammed, Evaluation the effect of lime on the plastic and
hardened properties of cement mortar and quantified using Vipulanandan model,
Open Engineering 9 (1) (2019) 468–480.
[30] C. Vipulanandan, A. Mohammed, Magnetic field strength and temperature effects
on the behavior of oil well cement slurry modified with iron oxide nanoparticles
and quantified with vipulanandan models, J. Test. Eval. 48 (6) (2019) 4516–4537.
[31] J.M. Marangu, Prediction of compressive strength of calcined clay based cement
mortars using support vector machine and artificial neural network techniques,
Journal of Sustainable Construction Materials and Technologies. 5 (1) (2020)
392–398.
[32] R. Ali, M. Muayad, A.S. Mohammed, P.G. Asteris, Analysis and prediction of the
effect of Nanosilica on the compressive strength of concrete with different mix
proportions and specimen sizes using various numerical approaches, Struct. Concr.
(2022).

24
[33] A.K. Ibrahim, H.Y. Dhahir, A.S. Mohammed, H.A. Omar, A.H. Sedo, The
effectiveness of surrogate models in predicting the long-term behavior of varying
compressive strength ranges of recycled concrete aggregate for a variety of shapes
and sizes of specimens, Archives of Civil and Mechanical Engineering 23 (1) (2023)
61.
[34] A. Ahmed, P. Abubakr, A. Salih Mohammed, Efficient models to evaluate the effect
of C3S, C2S, C3A, and C4AF contents on the long-term compressive strength of
cement paste, Structures 47 (2023) 1459–1475.
[35] Y. Sargam, K. Wang, I.H. Cho, Machine learning based prediction model for
thermal conductivity of concrete, Journal of Building Engineering 34 (2021),
101956.
[36] M.I. Shah, M.N. Amin, K. Khan, M.S.K. Niazi, F. Aslam, R. Alyousef, M.F. Javed,
A. Mosavi, Performance evaluation of soft computing for modeling the strength
properties of waste substitute green concrete, Sustainability 13 (5) (2021) 2867.
[37] Y. Li, F.N.S. Hishamuddin, A.S. Mohammed, D.J. Armaghani, D.V. Ulrikh,
A. Dehghanbanadaki, A. Azizi, The effects of rock index tests on prediction of
tensile strength of granitic samples: a neuro-fuzzy intelligent system, Sustainability
13 (19) (2021) 10541.
[38] H.U. Ahmed, A.S. Mohammed, A.A. Mohammed, R.H. Faraj, T. Xie, Systematic
multiscale models to predict the compressive strength of fly ash-based geopolymer
concrete at various mixture proportions and curing regimes, PLoS One 16 (6)
(2021) e0253006.
[39] R.H. Faraj, A.A. Mohammed, A. Mohammed, K.M. Omer, H.U. Ahmed, Systematic
multiscale models to predict the compressive strength of self-compacting concretes
modified with nanosilica at different curing ages, Eng. Comput. 38 (Suppl 3)
(2022) 2365–2388.
[40] M.L. Wang, V. Ramakrishnan, Evaluation of blended cement, mortar and concrete
made from type III cement and kiln dust, Constr. Build. Mater. 4 (2) (1990) 78–85.
[41] B. Rai, S. Kumar, K. Satish, Effect of Fly Ash on Mortar Mixes with Quarry Dust as
Fine Aggregate, Adv. Mater. Sci. Eng. 2014 (2014) 1–7.
[42] S.S.M. Ghoneim, S.S. Dessouky, A.A. Elfaraskoury, A.B. Abou Sharaf, Modelling
and experimental verification of barrier effect on breakdown voltage of
transformer oil using Box-Behnken Design, Measurement 147 (2019), 106829.
[43] W. Emad, A. Salih, R. Kurda, Forecasting the mechanical properties of soilcrete
using various simulation approaches, Structures 34 (2021) 653–665.
[44] A.A. Abdalla, A. Salih Mohammed, Theoretical models to evaluate the effect of
SiO2 and CaO contents on the long-term compressive strength of cement mortar
modified with cement kiln dust (CKD), Archives of Civil and Mechanical
Engineering 22 (3) (2022) 105.
[45] M.H. Akeed, S. Qaidi, H.U. Ahmed, W. Emad, R.H. Faraj, A.S. Mohammed, A.
R. Azevedo, Ultra-high-performance fiber-reinforced concrete. Part III: Fresh and
hardened properties, Case Studies Constr. Mater. 17 (2022) e01265.
[46] N.S. Piro, A.S. Mohammed, S.M. Hamad, The impact of GGBS and ferrous on the
flow of electrical current and compressive strength of concrete, Constr. Build.
Mater. 349 (2022), 128639.
[47] A.S. Al-Harthy, R. Taha, F. Al-Maamary, Effect of cement kiln dust (CKD) on mortar
and concrete mixtures, Constr. Build. Mater. 17 (5) (2003) 353–360.
[48] A. Abdalla, A.S. Mohammed, Hybrid MARS-, MEP-, and ANN-based prediction for
modeling the compressive strength of cement mortar with various sand size and
clay mineral metakaolin content, Archives of Civil and Mechanical Engineering 22
(4) (2022) 1–16.
[49] A. Mohammed, S. Rafiq, P. Sihag, R. Kurda, W. Mahmood, Soft computing
techniques: systematic multiscale models to predict the compressive strength of
HVFA concrete based on mix proportions and curing times, Journal of Building
Engineering. 33 (2021), 101851.
[50] I.A. Sharaky, S.S.M. Ghoneim, B.H. Abdel Aziz, M. Emara, Experimental and
theoretical study on the compressive strength of the high strength concrete
incorporating steel fiber and metakaolin, Structures 31 (2021) 57–67.
[51] D.J. Armaghani, G.D. Hatzigeorgiou, C. Karamani, A. Skentou, I. Zoumpoulaki, P.
G. Asteris, Soft computing-based techniques for concrete beams shear strength,
Procedia Struct. Integrity 17 (2019) 924–933.
[52] A. Kooshkaki, H. Eskandari-Naddaf, Effect of porosity on predicting compressive
and flexural strength of cement mortar containing micro and nano-silica by multi-
objective ANN modeling, Constr. Build. Mater. 212 (2019) 176–191.
[53] Asteris, P. G., Apostolopoulou, M., Armaghani, D. J., Cavaleri, L., Chountalas, A. T.,
Guney, D., ... & Nguyen, H. (2020). On the metaheuristic models for the prediction
of cement-metakaolin mortars compressive strength. 1, 1(1), 063.
[54] E.M. Golafshani, A. Behnood, M. Arashpour, Predicting the compressive strength of
normal and high-performance concretes using ANN and ANFIS hybridized with
Grey Wolf Optimizer, Constr. Build. Mater. 232 (2020), 117266.
[55] H.U. Ahmed, A.A. Abdalla, A.S. Mohammed, A.A. Mohammed, Mathematical
modeling techniques to predict the compressive strength of high-strength concrete
incorporated metakaolin with multiple mix proportions, Cleaner Materials 5
(2022), 100132.
[56] A. Abdalla, A. Salih, Microstructure and chemical characterizations with soft
computing models to evaluate the influence of calcium oxide and silicon dioxide in
the fly ash and cement kiln dust on the compressive strength of cement mortar,
Resources, Conservation & Recycling Advances 15 (2022), 200090.
[57] D.K.I. Jaf, P.I. Abdulrahman, A.S. Abdulrahman, A.S. Mohammed, Effitioned soft
computing models to evaluate the impact of silicon dioxide (SiO2) to calcium oxide
(CaO) ratio in fly ash on the compressive strength of concrete. Journal of Building,
Engineering 106820 (2023).
[58] H. Song, A. Ahmad, F. Farooq, K.A. Ostrowski, M. Maślak, S. Czarnecki, F. Aslam,
Predicting the compressive strength of concrete with fly ash admixture using
machine learning algorithms, Constr. Build. Mater. 308 (2021), 125021.
[59] T.R. Naik, S.S. Singh, B.W. Ramme, Effect of Source of Fly Ash on Abrasion
Resistance of Concrete, J. Appl. Sci. 14 (5) (2002) 417–426.
[60] L.H. Jiang, V.M. Malhotra, Reduction in water demand of non-air-entrained
concrete incorporating large volumes of fly ash, Cem. Concr. Res. 30 (11) (2000)
1785–1789.
[61] N. Bouzoubaâ, M. Lachemi, Self-compacting concrete incorporating high volumes
of class F fly ash preliminary results, Cem. Concr. Res. 31 (3) (2001) 413–420.
[62] C. Duran Atiş, Strength properties of high-volume fly ash roller compacted and
workable concrete, and influence of curing condition, Cem. Concr. Res. 35 (6)
(2005) 1112–1121.
[63] C.D. Atis, High-Volume Fly Ash Concrete with High Strength and Low Drying
Shrinkage, J. Mater. Civ. Eng. 15 (2) (2003) 153–156.
[64] N. Bouzoubaa, B. Fournier, Optimization of fly ash content in concrete: Part I: non-
air-entrained concrete made without superplasticizer, Cem. Concr. Res. 33 (7)
(2003) 1029–1037.
[65] G. Baert, A.-M. Poppe, N. De Belie, Strength and durability of high-volume fly ash
concrete, Struct. Concr. 9 (2) (2008) 101–108.

Machine learning techniques and multi-scale models to evaluate the impact of silicon dioxide (SiO2) and calcium oxide (CaO) in fly ash on the compressive strength of green concrete.pdf

More Related Content

Similar to Machine learning techniques and multi-scale models to evaluate the impact of silicon dioxide (SiO2) and calcium oxide (CaO) in fly ash on the compressive strength of green concrete.pdf (20)

More from Shakerqaidi (20)

Recently uploaded (20)

Machine learning techniques and multi-scale models to evaluate the impact of silicon dioxide (SiO2) and calcium oxide (CaO) in fly ash on the compressive strength of green concrete.pdf