Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

https://doi.org/10.1007/s41939-024-00619-9

ORIGINAL PAPER

Predictive modeling for compressive strength of blended cement

concrete using hybrid machine learning models
Asad Ullah Khan1 · Raheel Asghar1 · Najmul Hassan2 · Majid Khan3 · Muhammad Faisal Javed4,5 ·
Nashwan Adnan Othman6,7 · Shirin Shomurotova8

Received: 2 June 2024 / Accepted: 14 September 2024

© The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024

Abstract
Blended cement concrete, which incorporates supplementary cementitious materials alongside cement as a binder, is widely
recognized as a sustainable solution in the construction industry. However, accurate prediction of its properties can be
challenging due to its complex composition and numerous design parameters. Machine learning algorithms are commonly
employed to streamline this process, saving both time and resources. In this study, a novel approach was undertaken by utilizing
hybrid models that integrate support vector regression with optimization techniques, including the firefly algorithm, particle
swarm optimizer, and grey wolf optimizer, to predict the compressive strength of blended cement concrete. A comprehensive
dataset comprising 1287 data points, and 11 input variables was used for model training and validation. The hybrid models
demonstrated superior performance, with mean absolute error values below 6 MPa and root mean square error values below
9 MPa for both training and validation phases. Additionally, the correlation coefficient values for all models exceeded
0.87, indicating strong predictive accuracy. SHAP analysis revealed that the water-to-binder ratio was the most significant
factor influencing compressive strength, with a SHAP value of 9.8. The findings suggest that these hybrid machine-learning
models offer an effective tool for optimizing the compressive strength prediction of blended cement concrete in sustainable
construction practices, ensuring both resource efficiency and reliability. Future research could explore alternative machine
learning algorithms or novel hybrid combinations to further enhance model performance. Investigating these approaches may
address current limitations and potentially lead to more accurate and efficient predictions of the compressive strength of
blended cement concrete.

Keywords Blended cement concrete · Compressive strength prediction · Support vector regression · Optimization techniques ·
Sustainable construction

Abbreviations
B Majid Khan
[email protected] ANN Artificial neural network
B Binders
1 Department of Civil Engineering, COMSATS University BA Bottom ash
Islamabad, Abbottabad Campus, Abbottabad 22060, Pakistan BCC Blended cement concrete
2 Department of Transportation, School of Civil and C Cement
Environmental Engineering, National University of Science CA Coarse aggregate
and Technology (NUST), Islamabad, Pakistan
CaOH2 Calcium hydroxide
3 Department of Civil Engineering, Southern Illinois University CC Calcined clay
Edwardsville, Edwardsville, IL 62026, USA
4 Department of Civil Engineering, GIK Institute of
Engineering Sciences and Technology, Swabi 23640, Pakistan 7 Department of Computer Engineering, Al-Kitab University,
5 Department of Civil Engineering, College of Engineering in Altun Kupri, Iraq
Al-Kharj, Western Caspian University, Baku, Azerbaijan 8 Department of Chemistry Teaching Methods, Tashkent State
6 Department of Computer Engineering, College of Pedagogical University Named After Nizami, Bunyodkor
Engineering, Knowledge University, Erbil 44001, Iraq Street 27, Tashkent, Uzbekistan

0123456789().: V,-vol 123

 25 Page 2 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

CO2 Carbon dioxide φ(x) Function mapping the input data into a higher-
CS Compressive strength dimensional space
DT Decision tree
F Fly ash
FA Fine aggregate
FFA Firefly algorithm
GEP Genetic expression programming 1 Introduction
GGBS Ground granulated blast furnace slag
GWO Grey wolf optimizer According to recent World Bank reports, low-income coun-
LP Lime powder tries are experiencing the most rapid population growth,
MAE Mean absolute error accounting for approximately 40% of the world population
ML Machine learning (Juenger et al. 2019). The surge in population is accompa-
MRA Multiple regression analysis nied by a 65% rise in urbanization in these nations, driven
OPC Ordinary portland cement by urban policies that encourage this social trend (Miller
PSO Particle swarm optimizer et al. 2018). As the world’s most widely utilized construction
SCC Self-compacting concrete material (Asghar et al. 2020a, b), concrete is expected to see
SCMs Supplementary cementitious materials a significant rise in usage in low-income communities. The
SD Standard deviation current annual consumption of 15 billion tones is expected
SF Silica fume to at least double by 2050 (Akan et al. 2017). The significant
SHAP Shapley additive explanations amount of concrete production releases adverse emissions
SiO2 Silicon dioxide such as carbon dioxide (CO2 ), dust, and other pollutants,
SP Superplasticizers causing environmental pollution (Flower and Sanjayan 2007;
SVM Support vector machine Turner and Collins 2013; Thomas et al. 2009). On the other
SVR Support vector regressor hand, the growth of industrial activity has led to a continu-
W Water ous rise in waste materials. Consequently, waste management
W/B Water-to-binder ratio and proper disposal have emerged as significant concerns
A Control parameter (Ozcan et al. 2017). Fortunately, most of the waste materials
B Control parameter have cementitious properties (Yesilyurt 2018). Thus, utiliz-
OF Objective function ing waste as supplementary cementitious materials (SCMs)
Pk Best solution can help address both the environmental concerns connected
R or r Pearson correlation coefficient with waste disposal and the impact of cement on the envi-
RMSE Root mean square error ronment (Asghar et al. 2023; Amjad et al. 1098).
RRMSE Relative root mean square error Blended cement concrete (BCC) is a concrete that replaces
RSE Relative squared error a portion of ordinary Portland cement (OPC) with SCMs,
T Validation subset known as pozzolanic materials. Increasing the dosage of
Tmax Highest iteration count SCMs is anticipated to improve the sustainability of the con-
V Training subset crete product (Miller 2018). Currently, cements in the market
Vk Flight speed typically consist of approximately 20% SCMs (Fitriani et al.
b Bias term 2022). Excluding lesser-known and commercially limited
d Distance between fireflies SCMs like glass slag and bauxite residue, this paper focuses
e Actual value on widely utilized SCMs such as fly ash (F), ground granu-
ē Mean actual values lated blast furnace slag (GGBS), silica fume (SF), calcined
e−γdij2 Attractiveness between fireflies clay (CC), and lime powder (LP). Fly ash is a byproduct
f(x) Mathematical predictive function of coal burning in power plants and is abundantly produced
m Model-predicted value worldwide (Amran et al. 2020; Bhatt et al. 2019). Its utiliza-
m Mean model-predicted values tion in concrete has gained prominence due to its pozzolanic
n Dataset size properties (Khan et al. 2022). SF, also known as micro sil-
rand Random value from 0 to 1 ica, is a highly reactive pozzolanic material (Mazloom et al.
t Iteration count 2004). It is an amorphous form of silicon dioxide (SiO2 ) pro-
w Weight vector duced as a byproduct in manufacturing ferrosilicon alloys in
x Position of firefly or particle electric arc furnaces. GGBS is a byproduct generated by pro-
α Randomization factor ducing pig iron within blast furnaces (Özbay et al. 2016). It
β0 Scaling factor primarily consists of alumina and calcium-silicates, together

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 3 of 28 25

Fig. 1 The importance of SCMs

in producing environmentally
friendly concrete
Less CO2
Sustainable Waste material
emissions
construction recycling

Eco-friendly Supplementary
Enhance
materials cementitious
service life
materials

Improved
material Natural resources
Durability
properties prevention
aspect

with other bases. The slag undergoes rapid cooling through dominant chemical reaction (Hedayatinia et al. 2019). This
contact with water within a pond or through forceful water reaction occurs when the pore solution’s elevated pH level
jets, transforming it into a fine, granular form. This granu- (> 12) dissolves the passive anhydrous layer on F, releasing
lated slag exhibits a nearly entirely nanocrystalline, glassy aluminum and calcium ions in the solution. Subsequently,
structure and possesses latent hydraulic properties (Pal et al. these ions react with the calcium hydroxide (CaOH2 ) result-
2003). Furthermore, LP is a finely textured substance formed ing in cement hydration, resulting in expanded volume, and
by slaking quicklime with water, while CC is a sophisticated showing increased strength compared to CaOH2 (Lothen-
substance manufactured through the high-temperature calci- bach et al. 2011). It is important to note that SCMs like
nation of clay (Cardinaud et al. 2021). Additionally, Fig. 1 F in the binder mix delay the initial setting time of OPC,
displays the significance of SCMs in making environmen- consequently reducing the early age strength. Nevertheless,
tally friendly concrete. when up to 30% of OPC is replaced, the concrete’s strength
Compressive strength (CS) serves as a crucial gauge of displays a slight increase (< 10%) at 28 days of age and a
a concrete mix’s mechanical performance. It is considered more substantial increase (> 30%) at 90 days by replace-
the most representative measure, and other strength features, ment of up to 70% (Panesar and Zhang 2020). On the other
including elastic modulus and tensile strength, can be directly hand, supplementary materials like SF and CC contributed to
correlated with it for design. CS is typically determined the development of strength-bearing calcium silicate hydrate
using the standard test (ASTM C-39) after a curing period of through a pozzolanic reaction. Although the mechanism dif-
28 days (Kurda et al. 2019). Likewise, to other brittle mate- fers from that of F and GGBS, both SF and CC possess
rials, the primary determinant factor of concrete’s CS is its extremely fine particle sizes that render them highly reac-
porosity (Salami et al. 2021). Water-to-binder (W/B) ratio tive. This reactivity enables the compaction of the interfacial
holds a significant role in most concrete types (Al-Shamiri transition zone in the matrix, resulting in a very early set-
et al. 2019). By utilizing superplasticizers (SP) to reduce the ting and improved early strength compared to the mixture
W/B ratio while maintaining a consistent slump, it is possible with F (Scrivener et al. 2018). As a result, blended concrete
to increase the overall strength (Sathyan and Anand 2019). is expected to demonstrate strength increases of up to 40%
In high-strength concrete, aggregate quality becomes the pri- at 90 days (Sui et al. 2019). Regarding limestone powder,
mary governing parameter affecting strength (Einsfeld and its larger surface area, greater than that of GGBS and F but
Velasco 2006). When SCMs replace OPC, pozzolanic is the smaller than CC and SF, facilitates increased hydration of

123
 25 Page 4 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

OPC, leading to enhanced strength of concrete. Neverthe- methods proves valuable in estimating the strength charac-
less, due to its constrained pozzolanic activity as an SCM, teristics of concrete (Alabduljabbar et al. 2023; Chen et al.
LP contributes only minor strength improvements (< 15%) 2023; Ehteram et al. 2022; Azareh et al. 2021). Addition-
when replacing approximately 10–15% of OPC (Juenger ally, Fig. 2 presents a scientometric review highlighting the
et al. 2019). Extensive research has focused on investigat- significance of ML in cementitious composites.
ing the influence of using SCMs to restrict the content of Researchers have utilized various empirical and ML
cement and identify the optimum percentage for a replace- techniques to forecast concrete CS. ML models offer the
ment for concrete’s CS. However, it should be noted that advantage of incorporating multiple parameters and diverse
higher concentrations of SCMs can have detrimental effects data from concrete samples, surpassing the limitations of
on the composite matrix. Additionally, prompt availability empirical formulas. These methodologies establish a robust
of experimental test data is crucial for cost and time effi- framework for CS prediction in concrete. Furthermore,
ciency (Song et al. 2021). On-site and laboratory experiments hybrid and ensemble models are two approaches utilized
require sophisticated equipment, resulting in significant labor in ML to enhance the precision and performance of predic-
requirements and economic drawbacks (Nafees et al. 2021; tive models. Ensemble learning is a widely utilized method
Deng et al. 2024). The need for designated space to cure and that trains several weak learners independently, using train-
store concrete specimens during experiments also increases ing datasets and combining them to create robust learners.
costs. Moreover, considerable time is necessary to investi- On the other hand, hybrid models combine different mod-
gate the influence of various factors, incorporating concrete els or algorithms to create a unified model that capitalizes
mixture, aggregate constituents, and curing duration on CS on the strengths of each component. However, while artifi-
(Ahmad et al. 2021a). cial intelligence algorithms have been extensively utilized to
In addition to mitigating environmental impacts, the use predict the CS of concrete, an abundance of studies focuses
of SCMs introduces variability in concrete properties, mak- on the CS of SCMs containing concrete. The choice of novel
ing it essential to adopt advanced predictive methods, such as hybrid intelligent optimization requires further enhancement.
machine learning (ML) (Tipu et al. 2024a, 2023a, b, 2024b; Figure 3 shows the comparison of individual and hybrid mod-
2023c), to accurately forecast the CS of these sustainable con- els.
crete mixes. Numerous statistical and empirical relationships In this research, one individual SVR and three SVR-based
have been proposed to evaluate CS under different labora- hybrid models, including firefly algorithm (FFA), particle
tory conditions to overcome these challenges. However, these swarm optimizer (PSO), and grey wolf optimizer (GWO),
relationships are primarily developed for calculating CS in were used for predicting the CS of BCC. SVR was chosen
conventional concrete and cannot be relied upon for other for this study due to its strong ability to handle non-linear
concrete forms. Furthermore, these equations do not encom- relationships and high-dimensional data, which are common
pass the influence of all relevant factors, leading to reduced characteristics in predicting the compressive strength of BCC
estimation accuracy (Ahmad et al. 2022). The emergence of mixes. Unlike traditional regression models, SVR is built on
ML algorithms has provided a promising avenue to address the principles of structural risk minimization and is effective
these challenges and enhance research outcomes (Javed et al. in avoiding overfitting, making it a reliable tool for achieving
2024a; Asghar et al. 2024a, b, 2022; Asghar and Khattak accurate predictions even with limited data points. Moreover,
2023). Various ML algorithms, such as decision tree (DT), SVR’s ability to utilize various kernel functions enables it to
support vector regressor (SVR), genetic expression program- model complex relationships between input variables and the
ming (GEP), and artificial neural network (ANN), have been target output, providing better generalization and prediction
employed to address concrete-related challenges. Siddique accuracy compared to other machine learning algorithms.
et al. (2011) utilized ANN to accurately predict the CS of The FFA was selected for this study because of its simplic-
self-compacting concrete (SCC) incorporating bottom ash ity and effectiveness in global optimization problems. FFA
(BA), aiding in optimizing mix design. Similarly, Chou et al. mimics the behavior of fireflies, where less bright fireflies are
(2014) explored the forecasting abilities of ANN and SVR for attracted to brighter ones, allowing the algorithm to explore
estimating CS, enhancing accuracy and reliability in concrete and exploit the search space efficiently. This characteristic
structure design. Furthermore, Omran et al. (2016) evaluated makes FFA particularly suitable for optimizing the parame-
various data-mining approaches for forecasting the strength ters of SVR, as it balances exploration and exploitation, thus
of sustainable concrete. Chithra et al. (2016) utilized multi- preventing the model from getting stuck in local minima and
ple regression analysis (MRA) and ANN to predict the CS enhancing its predictive performance. PSO was included as
of SCC incorporating nano-silica and copper slag as partial an optimization technique because of its robust convergence
replacements. The study concluded that ANN outperformed properties and ease of implementation. PSO is inspired by the
MRA in terms of accuracy. Therefore, the application of ML social behavior of birds flocking or fish schooling, where par-
ticles (potential solutions) move through the problem space,

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 5 of 28 25

Fig. 2 Importance of ML approaches in concrete research

Individual learning model Hybrid learning models

Machine learning
algorithms
● Less stable model ● More stable model
● High bias and variance ● Low bias and variance
● Moderate performance ● Better performance
● Moderate prediction ● Better prediction
● Moderate to high errors ● Moderate to high errors

Fig. 3 Comparison of individual and hybrid algorithms

sharing information and adjusting their positions based on SVR model parameters in this study. The effectiveness of
their own experience and that of their neighbors. This collec- GWO in handling complex, multi-modal optimization prob-
tive intelligence helps PSO quickly converge to an optimal or lems further justifies its selection as one of the optimization
near-optimal solution, making it effective in fine-tuning the techniques.
hyperparameters of the SVR model, ensuring that the model This research uses three hybrid models based on SVR
captures the most relevant features for accurate predictions. to forecast the CS of SCMs concrete accurately. The exten-
GWO was chosen due to its ability to maintain a good balance sive database of 1287 samples is collected from the literature
between exploration and exploitation during the optimization (Hafez et al. 2022) to train and validate the models prop-
process. GWO mimics the leadership hierarchy and hunt- erly. These models are assessed using different evaluation
ing mechanism of grey wolves in nature, which allows it to criteria by incorporating the Taylor diagram. By consider-
explore the search space extensively before focusing on the ing eleven parameters extracted from concrete samples, our
most promising areas. This behavior helps to avoid prema- study aims to enhance the suitability of the suggested model
ture convergence and improves the global search capability and provide valuable insights for predicting the CS of SCMs
of the algorithm, making it well-suited for optimizing the containing concrete. The suggested models undergo external

123
 25 Page 6 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

validation assessments to assess their precision and reliabil-

ity using criteria suggested in existing research. Additionally,
the implementation of SHAP analysis helps determine the
importance and impact of input features on the concrete CS.
The study’s findings can serve as a practical guide for CS
prediction in SCM concrete applications.
This study introduces several innovative aspects and
makes significant contributions to the existing body of knowl-
edge on concrete CS prediction, particularly in the context
of BCC with SCMs. Unlike most other studies that focus on
conventional ML models or simple empirical relationships,
this research proposes three hybrid models based on SVR to
predict the CS of concrete with SCMs. These hybrid mod-
els integrate the strengths of different algorithms, offering
superior performance and accuracy in predicting CS com-
pared to traditional methods. The implementation of these Fig. 4 Illustration of SVR model decision boundary
hybrid models represents a novel approach in the field, par-
ticularly for SCM BCC, which has been less explored in
existing studies. The study leverages an extensive database of 2 Overview of ML approaches
1287 samples collected from existing literature (Hafez et al.
2022), encompassing a wide range of concrete compositions 2.1 Support vector regression
and conditions. This large and diverse dataset ensures that the
models are trained and validated comprehensively, enhancing SVR is a type of support vector machine (SVM) designed
their generalizability and applicability across different sce- for regression tasks (Smola and Schölkopf 2004). Unlike tra-
narios. Previous studies often relied on smaller, less diverse ditional regression models that minimize the error between
datasets, which limits the robustness of their predictive mod- predicted and actual values, SVR aims to fit the best line
els. The use of SHAP analysis in this study is an innovative within a predefined margin of tolerance, called epsilon.
approach to understanding the impact of various input fea- It uses kernel functions, such as linear, polynomial, and
tures on the predicted CS. By providing a detailed analysis radial basis functions, to transform input data into a higher-
of feature importance, this study not only improves the inter- dimensional space where a linear regression line can be
pretability of the ML models but also offers valuable insights fitted. SVR seeks to balance model complexity and prediction
into the key factors influencing concrete strength, which can accuracy by minimizing the error within the margin while
guide the design of more effective concrete mixtures. The penalizing deviations outside it, as given in Fig. 4 (Basak
proposed models undergo rigorous evaluation using multi- et al. 2007). This approach makes SVR effective for handling
ple criteria, including the Taylor diagram and are further non-linear relationships, preventing overfitting, and ensuring
validated through external assessments. This comprehen- robust predictions even with complex data (Burges 1998).
sive evaluation framework ensures that the models are not SVR aims to predict continuous outcomes, in the form of
only accurate but also reliable and robust across different a function i.e., f (x) that deviates from the actual observed
contexts. This level of thorough validation is a significant values by a small margin and is as flat as possible. Mathe-
advancement over existing studies that may not employ such matically it is given in Eq. (1). Here w is the weight vector,
detailed evaluation methods. The study’s findings have prac- φ(x) is a function that maps the input data into a higher-
tical implications for the construction industry, particularly in dimensional space (feature mapping), and b is the bias term.
the context of sustainable development. By accurately pre- SVR balances the complexity of the model (flatness) with the
dicting the CS of SCM-containing concrete, the proposed training error, making it robust to overfitting. By adjusting
models can facilitate the efficient use of SCMs, reducing the the model parameters, one can control the trade-off between
reliance on OPC and thereby mitigating the environmental achieving a low error on the training data and minimizing
impact associated with cement production. This contribution the model’s complexity. The performance of SVR heavily
is particularly relevant in light of the global push towards depends on the choice of kernel and its parameters. Proper
more sustainable construction practices. Overall, this study kernel selection and parameter tuning are crucial for captur-
bridges critical gaps in the literature by offering advanced ing the underlying data patterns effectively. While SVR with
predictive models that are both innovative and practically linear kernels is computationally efficient, using non-linear
applicable, particularly in the domain of sustainable concrete kernels increases the complexity and computational cost. For
construction. large datasets, this can be a limitation. SVR is particularly

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 7 of 28 25

Fig. 5 Flowchart of FFA

Start with FFA parameters and generate initial

population of fireflies

Assessment of fireflies fitness utilizing objective function

Updating the light intensity of fireflies

Ranking the fireflies and selecting the best

No
Moving the fireflies to their best solution

Iteration = max
generation?

Yes

Optimal outcomes

123
 25 Page 8 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

useful in situations where the relationship between variables

is complex and non-linear.

f (x)  wϕ(x) + b (1)

2.2 Firefly algorithm

The FFA is an algorithm based on principles from nature that

take inspiration from social behavior and flashing patterns of
fireflies. The algorithm operates based on the following three
rules, as described in Yang (2009):

• In the FFA, fireflies are drawn to one another based on

their brightness, irrespective of gender.
• The brightness of a firefly directly determines its attrac-
tiveness, which decreases as the distance between fireflies
increases. If a firefly cannot locate a brighter firefly nearby,
it will resort to random exploration within its neighbor-
hood.
• The objective function (OF) determines the brightness of
a firefly specific to the problem being solved.

The movement of a firefly towards another firefly with

higher brightness can be described using the expression (2)
(Yang and Karamanoglu 2013). In this equation, xi t and xi t+1
indicate the firefly’s current and new positions, respectively.
β0 represents a scaling factor, e−γd ij 2 represents the attrac-
tiveness between fireflies based on the distance (d) between
them, xj t is the position of the firefly with higher brightness,
α is a randomization factor, and rand shows a random value
ranging from 0 and 1. The flowchart of FFA is shown in
Fig. 5.
 
−γd2ij
i  xi + βo e
xt+1 xtj − xti + α(rand − 0.5)
t
(2)

2.3 Particle swarm optimizer

Kennedy and Eberhart (1995) introduced PSO, a technique of Fig. 6 Flowchart of PSO
evolutionary computation based on populations. This method
is commonly used for global optimization problems and emu-
lates the collective behavior observed in swarming animals rand2 are constants. Pk id denotes the optimal solution dis-
such as insects and bird flocks. In PSO, particles exchange covered by the particle, and Pk kk represents the best solution
information to update their positions to find the optimal solu- found by the entire particle population. PSO is a population-
tion in a D-dimensional space. Let Xk i represent the position based optimization technique inspired by the social behavior
of the ith particle after k iterations, Vk i denote the flight of bird flocking or fish schooling; the flowchart is shown in
speed, Pk i indicates the best solution found by the particle, Fig. 6.
and Pk represent the best solution found by the entire particle  
population. The speed and position updates of the particles Vk+1
id  wV k+1
id + c 1 rand 1 P k
id − X k
id
are performed iteratively based on the Eq. (3). where i  1, 2,  
…, m and d  1, 2, …, D. The variables w, c1 , c2 , rand1, and + c2 rand2 Pkkk − Xkid Xk+1 idi  Xid + Vid
k k+1
(3)

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 9 of 28 25

Fig. 7 Social hierarchy of grey wolves (Huang et al. 2023)

2.4 Grey wolf optimizer

The GWO, an optimization technique inspired by grey

wolves’ social structure and hunting behaviors, is based on
population intelligence (Mirjalili et al. 2014). In their natural
habitat, grey wolves exhibit a well-defined social structure
characterized by four distinct social groups. The relation-
ship between rank and the pursuit of dominance within the
grey wolves is illustrated in Fig. 7. Wolves employ a stalking
approach to locate their prey, followed by surrounding them
and gradually advancing towards it. The step-by-step pro-
cess of the GWO algorithm, a nature-inspired metaheuristic
optimization technique, is shown in Fig. 8.
The coordinated hunting nature of the wolf pack is
expressed in Eqs. (4)–(8). In the GWO algorithm, D rep-
resents the distance separating the grey wolf from the prey.
The existing iteration count is denoted by t, while Xp (t) and
X(t) represent the prey locations and the searching individ-
ual grey wolf after t iterations. Control parameters A and B
are utilized, and Tmax specifies the highest iteration count.
During the attack phase, the higher-ranking individuals α, β,
and δ take on the responsibility of locating the prey, guiding
the subordinate wolves ω through a series of approaching and
attacking behaviors. This coordinated effort aims to reach the
Fig. 8 Flowchart of GWO
aim of hunting prey. The calculation of parameters is given
in Eq. (9)–(11).
 
D  B.Xp (t) − X(t) (4)
B=2r2 (8)
X(t+1)  Xp (t) − AD (5)

⎧
a(t+1)  2(1 − t/Tmax ) (6) ⎪
⎨ Dα  |Bi .Xα − X|
Dβ  Bj .Xβ − X (9)
⎪
⎩ D  |B .X − X|
A=2ar1 − a (7) δ r δ

123
 25 Page 10 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

⎧
⎪
⎨ Xi  |Ai .Dα − Xα|
Marinković et al. 2017; Marques et al. 2013; Matos et al.
Xj  Aj .Dβ − Xβ  (10) 2019; McCarthy and Dhir 2005; McNally and Sheils 2012;
⎪
⎩ X  |A .D − X | Mittal et al. 2005; Moffatt et al. 2017; Mohammadi and South
r r δ δ
2017; Murad et al. 2019; Navarro et al. 2018; Nepomuceno
Xi + Xj + Xr et al. 2014; Nochaiya et al. 2010; Oner and Akyuz 2007; Oner
X(t+1)  (11)
3 et al. 2005; Park et al. 2012; Parron-Rubio et al. 2019; Patil
et al. 2013). The input parameters for this study include fine
aggregate (FA), water (W), and binders (B) which includ-
ing cement (C), F, LP, coarse aggregate (CA), GGBS, SF,
3 Methodology and calcinated clay. SCMs are used as substitutes for C in
varying proportions. Table 1 presents a comprehensive sta-
3.1 Data collection tistical overview of the input features and output response.
Its maximum and minimum values indicate the dataset’s
The database of 1287 samples is acquired from supplemen- extremes, while the mean, mode, and median provide insight
tary materials of published papers (Hafez et al. 2022), which into its central tendency. The standard deviation (SD) met-
were collected from literature (Fitriani et al. 2022; Eins- ric quantifies data variability, with a smaller SD suggesting
feld and Velasco 2006; Panesar and Zhang 2020; Meddah values closely grouped around the mean and a larger SD
et al. 2014; Felekoğlu et al. 2007; Teng et al. 2013; Megat indicating greater dispersion. Skewness and kurtosis assess
Johari et al. 2011; Poon et al. 2006; Wu et al. 2001; Divsholi distribution characteristics. Skewness measures asymmetry,
et al. 2014; Wongkeo et al. 2014; Wang 2019; Zhang et al. where zero indicates symmetry and positive/negative values
2018; Kou and Poon 2013; Preez and Phillip 2019; Quan and denote extensions in the right/left tails, respectively (Cain
Kasami 2014; Rathnarajan et al. 2017; Rozière et al. 2009; et al. 2017). Positive kurtosis and zero kurtoses characterize
Ruixia 2010; Saha et al. 2020; Şahmaran et al. 2009; Samad data with moderate peaks, resembling a normal distribution.
et al. 2017; San Nicolas et al. 2014; Shaikh and Supit 2015; Negative kurtosis indicates a flatter curve than the standard
Siddique 2004; Silva et al. 2013; Sonebi et al. 2008; Sout- distribution, while positive kurtosis indicates a more pro-
sos et al. 2018; Sugi et al. 2013; Sujjavanich et al. 2017; nounced peak (Benson 1993; Brown and Greene 2006). The
Tae et al. 2011; Uysal and Sumer 2011; Vejmelková et al. ranges considered acceptable for skewness and kurtosis are
2009, 2011; Vivek and Dhinakaran 2017; Vollpracht et al. from − 3 to + 3 and from − 10 to + 10, respectively (Brown
2018; Vu et al. 2001; Yazıcı 2008; Yeau and Kim 2005; and Greene 2006). Table 1 reveals that nearly all data points
Yoo et al. 2015; Zhao et al. 2015; Pillai et al. 2019; Kumar lie within the acceptable range, indicating a favorable distri-
et al. 2020; Heede et al. 2019, 2017; Panesar et al. 2017; bution pattern. In Fig. 9, the histogram distribution plots of
Chen et al. 2014; Inthata et al. 2013; Mohamed et al. 2018; each variable in the dataset providing a visual representation
Adam and Molyneaux 2009; Baghabra Al-Amoudi et al. of the data distribution are shown. They help to better under-
2009; Amankwah et al. 2015; Angulo-Ramirez et al. 2019; stand the spread, skewness, and outliers within the dataset.
Arora et al. 2019; Atiş 2003; Balakrishnan and Awal 2014; Each histogram displays the frequency of different value
Berndt 2009; Bilim et al. 2009; Biswas et al. 2017; Bucher ranges for the respective variables, offering insights into the
et al. 2017; Celik et al. 2015; Gholampour and Ozbakkaloglu central tendency, variability, and overall shape of the data.
2017; Czarnecki et al. 2018; Dhanya et al. 2018; Dhandapani This enables a clearer understanding of the underlying pat-
et al. 2018; Collepardi et al. 2004; Crouch et al. 2007; Diab terns and assists in assessing the suitability of the data for
et al. 2016; Dinakar et al. 2007, 2013; Duan et al. 2013; model development. Moreover, the distribution plots help
Durán-Herrera et al. 2015; Eguchi et al. 2013; Faleschini detect potential issues such as imbalanced or irregularly dis-
et al. 2015; Fanghui et al. 2015; Garcez et al. 2018; García- tributed data, ensuring that appropriate data preprocessing
Segura et al. 2014; Gesoğlu et al. 2009; Gettu et al. 2018; techniques are applied. The statistics of input features and
Golewski 2018; Güneyisi et al. 2010; Harrison et al. 2012; output response shown in Table 1 and histogram distribution
Hawileh et al. 2017; Holt et al. 2010; Shi et al. 2009; Hussain plots of each variable in the dataset shown in Fig. 9 reveal
et al. 2013; Jau et al. 2004; Jiang et al. 2004; Jones et al. 1997; the representativeness of the database, encompassing a wide
Kaewmanee and Tangtermsirikul 2014; Karahan 2017; Karri range of concrete compositions and conditions.
et al. 2019; Kavitha et al. 2016; Khodair and Bommareddy The interrelationship between input features is assessed
2017; Khodabakhshian et al. 2018; Kou et al. 2007, 2011; using the Pearson correlation coefficient (R or r). In Fig. 10,
Khunthongkeaw et al. 2006; Kurda et al. 2018; Leemann et al. the upper triangle depicts the calculated R values, the lower
2015; Lee et al. 2013; Leung et al. 2016; Lima et al. 2013; triangle displays scatterplots illustrating the dependencies
Limbachiya et al. 2012; Ling et al. 2004; Liu et al. 2014; Löf- between these values, and the diagonal exhibits the distri-
gren et al. 2016; Long et al. 2015, 2017; Lübeck et al. 2012; bution of the input features. The correlation among all input

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 11 of 28 25

Table 1 Various statistics of input features and output response

Parameter B (Kg/m3 ) C/B W/B GGBS/B F/B LP/B SF/B CA/B CC/B SP/B FA/B CS (MPa)

Mean 422.84 0.72 0.44 0.12 0.13 0.01 0.01 2.51 0.008 0.01 1.97 48.31
Kurtosis − 0.11 − 0.79 2.28 1.53 0.65 13.44 8.55 8.04 8.76 6.57 3.66 − 0.30
SD 95.49 0.22 0.12 0.21 0.18 0.05 0.03 1.04 0.04 0.07 0.69 18.99
Median 410 0.73 0.42 0 0 0 0 2.41 0 0 1.88 45
Maximum 770 1 1.19 0.9 0.9 0.47 0.15 21.38 0.4 1.15 6.185 110
Standard Error 2.66 0.006 0.003 0.006 0.005 0.002 0.001 0.03 0.001 0.002 0.02 0.53
Sample Variance 9117.92 0.05 0.02 0.04 0.03 0.003 0.001 1.07 0.002 0.005 0.47 360.89
Range 595 0.9 1 0.9 0.9 0.47 0.15 20.61 0.4 1.15 5.71 105
Skewness 0.25 − 0.40 1.18 1.66 1.24 3.25 3.06 3.10 3.89 1.69 1.41 0.53
Mode 400 1 0.4 0 0 0 0 2.5 0 0 2 40
Minimum 175 0.1 0.19 0 0 0 0 0.77 0 0 0.47 5

features is notably low, indicating the absence of collinearity Python scripts for enhancing model performance. These soft-
concerns. This meets the requirement outlined in the liter- ware and tools ensure the reproducibility of the study and
ature, where correlations are advised to remain below 0.8 provide transparency in the computational workflow. The
(Farooq et al. 2021). Moreover, the correlation between the objective of the study is to improve the predictive capacity
input variables and the CS is notably strong, indicating that of SVR in estimating CS within BCC. The vital parameters
all input features substantially influence CS. The W/B ratio of SVR, namely C and g, are both defined within the limits
(r  − 0.660) is the most influential feature, which exhibits of ranges from 0.02 to 100. For the enhancement of SVR’s
an inverse relationship with CS. This phenomenon can be predictive performance, three metaheuristic models, namely
attributed to the increase in W, which results in the forma- PSO, GWO, and FFA, are employed. These models undergo
tion of pores within the concrete. These pores have an adverse training using 901/1286 data sets and are validated using
effect on the concrete’s strength properties. The amount of B 387/1286 data sets, with a distribution of 70% for training and
(r  0.621), C/B (r  0.154), and SP/B (r  0.161) demon- 30% for validation. This split was chosen to ensure a robust
strate a direct relationship with CS. SP is utilized to enhance model training process while maintaining a sufficiently large
the workability of concrete mixtures by improving fluidity validation set to accurately assess model performance. The
and flow characteristics, obviating the requirement for addi- 70–30 split is a common practice in machine learning, as it
tional W. By reducing the water-cement ratio, these agents provides a balanced approach to training the model while
create more dense concrete, enhancing its CS. The contri- retaining enough data to evaluate its generalization ability
bution of SCMs is negative, as they possess lower binding effectively. The flowchart in Fig. 11 presents SVR-based
properties compared to cement, leading to weaker concrete. models optimized using GWO, FFA, and PSO, depicting a
coherent process where optimization techniques improve the
SVR algorithm to enhance predictive accuracy when estimat-
ing the CS of concrete incorporating SCMs. Comprehensive
3.2 Model development parameter configurations for GWO, FFA, and PSO are pro-
vided in Table 2, influencing fitness values across iterations.
The ML models in this study were developed using Python The final parameter settings for the FFA, PSO, and GWO
(version 3.9). The analysis utilized several key libraries and algorithms were determined based on values adopted from
packages essential for machine learning and data process- relevant literature (Javed et al. 2024b, c; Aldrees et al. 2024).
ing. The scikit-learn library (version 0.24.2) was used to These settings were selected after reviewing studies where
implement the SVR algorithm, while numpy (version 1.21.0) similar algorithms were applied successfully, ensuring opti-
was employed for numerical operations and pandas (version mal performance in the context of our research.
1.3.0) for data manipulation and analysis. Data visualiza-
tion was conducted using Matplotlib (version 3.4.2) and
seaborn (version 0.11.1). Additionally, the SHAP library 3.3 Performance evaluation
(version 0.39.0) was used for model interpretability. The
PSO, FFA, and GWO optimization algorithms were imple- The model’s efficacy is evaluated using statistical tech-
mented through specific optimization libraries or custom niques on two subsets (training and testing). These methods

123
 25 Page 12 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Fig. 9 Distribution histograms of each parameter in the dataset

include mean absolute error (MAE), relative squared error the degree of alignment between a curve and actual data. A
(RSE), root mean square error (RMSE), relative root mean score of 1 signifies an ideal match between actual and pre-
square error (RRMSE), and correlation coefficient (R). The dicted values. R values falling within the range of 0.8–0.87
RMSE handles larger error magnitudes. A lower RMSE value are considered acceptable, while values below 0.70 indi-
indicates that the constructed model has minimal error impli- cate unsatisfactory performance (Gandomi and Roke 2015;
cations and demonstrates superior performance (Despotovic Gandomi et al. 2011, 2012). In physical engineering, the a10-
et al. 2016). However, it has been noted that the MAE is index holds notable significance, and as it closes to 1.0, the
more proficient in addressing scenarios involving continu- quality of the prediction model is enhanced. In addition, the
ous and smooth data. The R is a widely used metric to assess OF was also employed for assessing model overfitting issues.

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 13 of 28 25

Fig. 10 Pearson correlation matrix and distribution of input parameters

Furthermore, the model’s external validation was conducted n

1 i=1 (ei − m i )2
using evaluation criteria recommended in the literature, as RRMSE  (15)
outlined in Table 3. These statistical evaluations can be math- |e| n
ematically represented using Eqs. (12)–(18). In this context, n
− ei )(m i − mi
i1 (ei
ei denotes the ith actual value, while ēi represents the mean of R (16)
n n 2
the actual values. Similarly, mi refers to the model-predicted i1 (ei − e)2 i=1 (m i − m)
value for the ith instance, and mi represents the mean of the RRMSE
model-predicted values. The symbol n denotes the overall ρ (17)
1+R
dataset size utilized in formulating the model, while sub-
scripts V and T indicate the training and validation subsets, ρT (nT − nv) 2ρV n v
OF  + (18)
respectively. n n
n
i=1 |ei − mi |
MAE  (12)
n 3.4 SHAP analysis

n
i1 (ei − m i )2 The Shapley additive explanations (SHAP) approach is a
RMSE  (13) powerful technique for explaining the output of ML models.
n
It is based on Shapley values from cooperative game theory
n
i=1 (m i − ei )
2
and provides a consistent way to attribute predictions to fea-
RSE  n (14)
i=1 (e − ei )
2 tures in a model (Lundberg 1705). SHAP values quantify the

123
 25 Page 14 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Database of SFRC compressive

GWO FFA PSO strength

Parameters Initialization Training set (70%) Testing set (30%)

Fitness calculation Initial parameters of

SVR model

No Reducing fitness function

(RMSE of each iteration) Satisfy a stop Optimal SVR
No Yes
condition? prediction model

Reach max
iteration Fitness assessment

Yes SVR-FFA
SVR-GWO
Output of the parameters SVR-PSO

Fig. 11 Flowchart of SVR-based hybrid model

Table 2 Parameter setting of metaheuristic algorithms Table 3 External validation metrics

Algorithm Parameter Value S. Formula limits References

no.
PSO c1 1.5
n
i1 (ei×mi)
c2 1.6 1 k ei2
0.85 < k < 1.15 Golbraikh and
Vmax 6 Tropsha
(2002)
wmax 0.8 n
(ei ×mi)
2 k  i=1
mi2
0.85 < k < 1.15 Golbraikh and
wmin 0.2
Tropsha
GWO a [0, 2] (2002)
FFA α 1 3 Rm >0.5 Roy and Roy
Rm  R2 × (1 −
β0 2 (2008)
|R2 − R2o |)
δ 0.2
rand 0.95 Where
R2  R2 ∼
1
n o 2
i1 (mi−ei )
1− n o )2
i1 (mi-mi
impact of each feature on the model’s output by considering
eoi  k×mi
all possible combinations of features and their contributions
R2o  R2o ∼
1
to the prediction (Wu and Zhou 2022, 2023). This enables n o 2
i=1 (ei-mi )
us to understand not only the importance of each feature 1- n o 2
i=1 (ei-ei )
but also the interactions between them. The SHAP approach
m io  k × ei
offers several benefits, including model agnosticism, mean-
ing it can be applied to any ML model regardless of its
complexity or architecture. It also provides both local and
global interpretability, allowing us to understand individual
predictions as well as overall model behavior. Overall, the
SHAP approach is a valuable tool for gaining insights into
ML models and improving transparency and trust in their
predictions, as given in Fig. 12.

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 15 of 28 25

Fig. 12 SHAP attribute flowchart

Table 4 Various statistical

parameters of models Model SVR SVR-PSO SVR-GWO SVR-FFA

Subset Training Validation Training Validation Training Validation Training Validation

MAE 6.94 8.37 6.25 7.55 5.66 6.91 4.60 5.54

RMSE 9.17 10.93 8.35 10.01 7.43 8.95 6.17 7.44
RSE 0.24 0.30 0.20 0.25 0.16 0.20 0.11 0.13
RRMSE 0.19 0.22 0.17 0.20 0.15 0.18 0.12 0.15
R 0.87 0.84 0.89 0.87 0.92 0.91 0.95 0.93
ρ 0.10 0.12 0.09 0.11 0.08 0.09 0.06 0.08
OF 0.11 0.19 0.09 0.08

MAE MAE
8 9.6

7 8.4
4
ρ RMSE ρ RMSE
10 7.2
7 2 11.7
6 0.14
0.12 8 10.4
4
0.12 6.0
6 0 Validation
0.10 Training 9.1
9
9.
5 6 10
1
0.100
0.088 SVR
0.06
0 066 SVR 0
0..0
08
0.08
SVR-PSO
0.126
0.122266 SVR-PSO 0.189
0.18
0 89 SVR-GWO
SVR-GWO SRV-FFA
0.189
0.189 0.864
0.86
0 864 0.252
0.2
25
0.900
0.90
.900000 0.125
0.125 SRV-FFA
0.252 0.88
0 88
0.888 0.175
0
0.17
.175
.17
17 0.315
0.925
09 5 0.912
0.950 0.150
0.1500 R 0.936
R RSE 0 0
0.200 RSE

0.175
0.175
0.225

0.200
RRMSE RRMSE
(a) (b)
Fig. 13 Radar diagram; a training data, b validation data

4 Results and discussion 6.94 MPa for training and 8.37 MPa for validation. Notably,
the hybrid models exhibit substantial improvements. Specif-
4.1 Statistical evaluation ically, the SVR-PSO hybrid reduces the MAE to 6.25 MPa
for training and 7.55 MPa for validation, while the SVR-
This work focused on assessing the efficacy of the SVR FFA hybrid achieves an even more significant reduction,
method to predict the CS of concrete and its enhancement with MAE values of 4.60 MPa for training and 5.54 MPa
by utilizing three different hybrid models. These models are for validation. In the training phase, the RMSE, RSE, and
statistically evaluated using different statistical parameters, RRMSE values for the SVR model are 9.17 MPa, 0.24,
including RSE, RMSE, MAE, RRMSE, ρ, and OF. Table and 0.19, respectively. During the validation phase, these
4 presents these values for both the training and the vali- values shift to 10.93 MPa for RMSE, 0.30 for RSE, and
dation. The MAE for the standalone SVR model stands at

123
 25 Page 16 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

0 0.2 below 10 MPa. In contrast, the hybrid models in this study

0.4 Reference
achieved superior accuracy, with the SVR-FFA model reach-

Co
SVR-PSO

rre
20 0.6 SVR ing a correlation of 0.95 (training) and 0.93 (validation), and

lat
SVR-GWO significantly lower MAE values of 4.60 MPa (training) and

io
Standard Deviation

nC
SVR-FFA 5.54 MPa (validation). Similarly, the study by Ahmad et al.
0.8

oe
15 (Ahmad et al. 2021b) reported a model with a correlation

ffi
cie
value of 0.92. The SVR-FFA model developed in this study
0.9

nt
10 surpasses this, demonstrating better correlation and predic-
0.95 tive capability. Another study by Karim et al. (2024) achieved
a correlation value of 0.84 based on a dataset containing
5
0.99
138 datapoints. This study achieved a higher correlation
value using a significantly larger dataset. In summary, the
0 1 hybrid models developed in this research demonstrate better
0 5 10 15 20 accuracy to existing benchmarks or studies, particularly in
Standard Deviation reducing error and improving correlation. The results place
these models among the most accurate and reliable methods
Fig. 14 Taylor diagram for predicting the CS of concrete.

4.2 Comparison of the slope of the regression line

0.22 for RRMSE. All three hybrid models result in a sub-
stantial decrease in these values. Furthermore, the R-value In this section, a comparative analysis is conducted by plot-
of the SVR is 0.87 for training and 0.84 for validation. ting original values on the x-axis against those predicted on
The SVR R-value for training exhibits a 2.75% increment the y-axis. A trend line is then fitted to all the data points. This
through SVR-PSO and a 5.5% increase through SVR-GWO, trend line should have one slope in an ideal scenario, signi-
and notably, a 9% enhancement is observed with the SVR- fying a perfect match between experimental and forecasted
FFA model. The OF values stand at 0.11 for the SVR and outcomes. As the slope approaches zero, the model’s accu-
0.08 for the SVR-FFA. These scores are notably small and racy increases. A value falling within the range of 0.8 to 1.2
approach zero, indicating the absence of overfitting concerns. suggests a strong and superior correlation between the exper-
The SVR model generally demonstrates commendable sta- imental and model results (Jalal et al. 2021a, b). Figure 15a
tistical accuracy, characterized by minimal error values and a compares the actual and SVR-predicted values. The slope
strong correlation. Introducing hybridization further elevates of the regression line is computed as 0.79 for the training
the model’s performance, with the SVR-FFA model display- dataset and 0.71 for the validation dataset of the SVR model.
ing the most remarkable results. The comparative analysis The utilization of the hybrid-SVR model further enhances
of these models is visually depicted in Fig. 13 through a these slopes toward unity. The slope of the regression line
radar diagram. Notably, the SVR-FFA model exhibits the becomes 0.81 for SVR-PSO, 0.82 for SVR-GWO, and 0.88
least error and the highest R-value, underscoring its supe- for SVR-FFA concerning the training datasets, as shown in
riority over the other hybrid models. The Taylor diagram Fig. 15b–d. Notably, regarding regression slopes, the SVR-
presented in Fig. 14 offers a visual representation of the pre- FFA model demonstrates higher accuracy than other hybrid
dictive model assessment. The radial lines correspond to the models.
model’s correlation coefficients, the red angular lines illus-
trate the RMSE values and the black radial lines depict the 4.3 Error analysis
SD. The blue reference point indicates the optimal point. In
this context, the SVR-FFA model shows enhanced accuracy Errors in ML predictive models typically arise from issues
by closely approximating the reference point. Conversely, like the bias-variance tradeoff, where a model may be too
the individual SVR model is positioned at a greater distance, simple or too complex. Insufficient or noisy data, poor fea-
underscoring its relative accuracy compared to hybrid mod- ture selection, and improper data preprocessing can also
els. lead to inaccuracies. Additionally, selecting an inappropriate
The results of this study demonstrate the high accuracy model or suboptimal hyperparameters can further contribute
in predicting the CS of concrete. When compared to exist- to errors. To reduce these errors, it is essential to ensure
ing studies, such as that by Aslam and Shahab (Aslam and high-quality and sufficient data, apply effective feature engi-
Shahab 2024), where various ML models were evaluated for neering, and carefully manage model complexity through
predicting CS, the best model achieved a correlation value regularization and external validation. Employing ensem-
of 0.89, with most models showing MAE and RMSE values ble methods, optimizing hyperparameters, and continuously

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 17 of 28 25

120
120 Training Training
Validation Validation
y=0.79x+12 (training) 100 y= 0.81x+11(training
100 y=0.74x+15 (validation

SVR-PSO predicted CS (MPa)

y=0.71x+17 (validation)
y=x (ideal fit line) y=x (ideal fit line)
SVR predicted CS (MPa)

80
80

60
60

40
40

20 20

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Actual CS (MPa) Actual CS (MPa)
(a) (b)
120
Training
120
Validation
100 y=0.82x+11 (training) Training
Validation
SVR-GWO predicted CS (MPa)

Y=0.73X+15 (validation) 100

y=x (ideal fit line) y=0.88x+8 (training)
SVR-FFA predicted CS (MPa)

80 y=0.82x+11 (validation)
y=x (ideal fit line)
80

60
60

40
40

20
20

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Actual CS (MPa) Actual CS (MPa)

(c) (d)

Fig. 15 Comparison of regression slope; a SVR model, b SVR-PSO model, c SVR-GWO model, and d SVR-FFA model

monitoring model performance are also crucial steps in min- further, as substantiated by reducing the average error to
imizing errors and improving prediction accuracy. 6.64 MPa through PSO-SVR, 6.04 MPa through GWO-SVR,
In Fig. 16, a graphical representation presents actual and, most significantly, down to 4.9 MPa through FFA-SVR.
outcomes, model results, and their corresponding absolute Figure 16c, d illustrates the graph and histogram of absolute
errors. Additionally, a histogram depicting the distribution of errors for the PSO-SVR model. The data shows a positive
absolute errors is included to provide insights into their sta- pattern, with 57% of errors registering values below 6 MPa
tistical characteristics. Figure 16a shows that the SVR model and a significant 88% remaining within the range of 12 MPa
closely tracks the actual results, yielding an average error of or lower. Similarly, the SVR-GWO model closely aligns with
7.47 MPa. The histogram distribution of absolute errors for experimental values, with only a 37% error rate exceeding
the SVR model is examined in Fig. 16b. Notably, approxi- 6 MPa. The progression of actual and model values for the
mately 52% of data points exhibit errors below 6 MPa, while SVR-FFA model is presented in Fig. 16g. Notably, 70% of
80% of errors remain under 12 MPa. A mere 20% of data errors are below 6 MPa, and a significant 90% remain within
points exhibit errors exceeding the threshold of 12 MPa. the range of 12 MPa or lower. The error analysis emphasizes
The efficacy of the individual SVR model is enhanced even that all model values closely correspond to the actual values;

123
 25 Page 18 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

140 120
Actual Predicted by SVR Frequency
Abs error Average error=7.37 400 394 Comulative %
120 100
Maximum error=33 Minimum error=0.001

100
300 80

Cumulative Percent
Frequency (SVR)
273
80
CS (MPa)

219 60
200
60

138 40
40 106
100
20
20 54
42
28 21
6 3 3
0 0 0
0 6 12 18 24 30 36
Absolute error
0 200 400 600 800 1000 1200 1400
Dataset
(a) (b)
140 500 120
Actual predicted by SVR-PSO Frequency
Abs error Average error=6.64 440 Comulative %
120
Maximum error=32.49 Minimum error=0.003
100
400
100

Frquency (SVR-PSO)

Cumulative Percent
80
80
300
CS (MPa)

287

60 60
202
200
40
143 40

20
100 91
20
50
0 33
19 13 8
0 200 400 600 800 1000 1200 1400 0 0
Dataset 0 6 12 18 24 30
Absolute error
(c) (d)

140 500 120

Actual Predicted by SVR-GWO Frequency
448 Comulative %
Abs error Average error=6.04
120
Maximun error=33.57 Minimum error= 0.008 100
400
100
Frequency (SVR-GWO)

Cumulative Percent
311 80
80 300
CS (MPa)

60
60 222
200
40 40
142

20 100 86
20
39
0 18 10 7 3
0 0
0 6 12 18 24 30
0 200 400 600 800 1000 1200 1400
Dataset Absolute error
(e) (f)
140 120
Actual Predicted by SVR-FFA 600 574 Frequency
Abs error Average error=4.9 Comulative %
120
Maximum error=24.27 Minimum error=0.008 100
500
100
Frequency (SVR-FFA)

80
Cumulative Percent

400
80
CS (MPa)

309 60
60 300

40 200 194 40

20 102
100 20
62
0 20 21
4
0 0
0 200 400 600 800 1000 1200 1400 0 6 12 18 24
Dataset Absolute error
(g) (h)

Fig. 16 Experimental vs. model values trend and error analysis for a, b SVR, c, d SVR-PSO, e, f SVR-GWO, g, h SVR-FFA

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 19 of 28 25

nonetheless, the SVR-FFA model outperforms other mod- SHAP summary plots are shown in Fig. 17b, which illus-
els by displaying lower absolute errors and a more precise trates the impact of each feature on the output response. The
alignment with the actual outcomes. vertical axis shows the input variables and their significance
levels, while the horizontal axis explains specific SHAP val-
4.4 External validation ues. The color variation of the data points indicates their size,
ranging from rare blue to darker red. The horizontal axis illus-
The proposed models underwent external validation assess- trates the influence of the features on forecast records using
ments to determine their accuracy and reliability, as shown SHAP values, progressing from the color blue to red as the
in Table 5. The SVR, SVR-PSO, SVR-GWO, and SVR-FFA intensity of features varies. It is evident that C/B, B, and SP
models display k and k’ values that are nearly equal to 1, positively affect the output response. Conversely, a reverse
meeting the criteria established by Golbraikh and Tropsha effect is observed for W/B, F/B, CA/B, and LP/B on the CS
(Golbraikh and Tropsha 2002). Additionally, the Rm val- of concrete.
ues for both models exceed the 0.5 threshold, adhering to To illustrate the influence of SHAP values on a particu-
the standards recommended by Roy (Golbraikh and Trop- lar predictive outcome, consider a specific example from the
sha 2002). Consequently, all models confirm their validity prediction of CS as shown in Fig. 18. In this instance, W/B
through robust performance, indicating that their effective- has the highest SHAP value, indicating it is the most influ-
ness is not merely due to a simplistic correlation between ential factor in determining CS. For a specific concrete mix,
input variables and the response parameter. the model predicts a lower-than-average CS, and the SHAP
The heftiness of hybrid models has been demonstrated in analysis reveals the reason for it. For this mix, the W/B ratio
several studies where they have successfully predicted var- has a negative SHAP value of 9.8, meaning it is significantly
ious concrete properties. For instance, Javed et. al. (Javed pushing the predicted value down. This suggests that a higher
et al. 2024b) effectively predicted the strength properties W/B ratio negatively impacts the CS, which is consistent with
of sustainable green concrete containing waste foundry the understanding that excess water weakens the concrete
sand, showcasing the potential of hybrid approaches. Simi- matrix. Alongside W/B, the F/B has a negative SHAP value
larly, metaheuristic algorithms were successfully applied for of 2.4, further lowering the predicted strength but with lesser
predicting compressive strength in steel fiber-reinforced con- impact. The C/B and B have a positive SHAP values of 2.2
crete exposed to high temperatures in another study (Javed and 1.9, respectively. This indicates that both features con-
et al. 2024c), highlighting the resilience of these models tribute positively to the predicted CS of the concrete, meaning
under extreme conditions. Furthermore, hybrid metaheuris- their increase leads to a higher predicted outcome for CS.
tic optimized SVR models were utilized for shear strength This example illustrates that, for this specific prediction, the
estimation of reinforced concrete deep beams (Kaloop et al. model’s outcome of lower CS is primarily driven by the high
2022), yielding accurate results. Other successful applica- W/B and F/B ratios, even though the C/B and B content
tions include studies conducted by Hameed et al. (2022) attempt to improve the outcome. The SHAP values clearly
and Seghier et al. (2023), which predicted the compressive explain how different features contribute to the model’s deci-
strength of concrete containing industrial waste materials and sion, offering transparency into how the mix design affects
modeled the compressive strength of concrete with waste the predicted CS of concrete.
glass. These instances demonstrate the capability of hybrid
models to provide reliable predictions across diverse concrete 4.6 Potential and complications of developed
compositions and conditions. models in practical applications

4.5 SHAP analysis The predictive models developed in this study offer immense
potential for real-world applications, particularly in the con-
Figure 17a illustrates the significance of input features on the struction industry, where sustainability and cost-efficiency
CS of concrete. The input features are arranged in descend- are of growing importance. Specifically, these models could
ing order. Among the inputs, the W/B emerges as the most be beneficial to industry partners such as construction com-
influential factor, with a SHAP value of 9.8, significantly panies, cement manufacturers, and civil engineering firms.
affecting the output response. Factors such as the F/B ratio Global companies that are leaders in producing cementi-
(SHAP  2.4), C/B ratio (SHAP  2.2), and B (SHAP  tious materials could leverage these models to optimize their
1.9) also exhibit significant influence on the output response. blended cement formulations and reduce reliance on OPC,
Moreover, parameters such as CA/B, SP/B, and SF/B exhibit thereby lowering carbon emissions. Additionally, infrastruc-
a moderate and comparable significance level. On the other ture projects that require durable and sustainable materials,
hand, attributes like CC/B, GGBS/B, and FA/B demonstrate such as highway construction, bridges, and large-scale hous-
a relatively lower contribution to the CS of concrete. The ing developments, could benefit from the use of these

123
 25 Page 20 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Table 5 External validation

metrics scores Parameter SVR model SVR-PSO SVR-GWO model SVR-FFA model

k 0.961 0.965 0.966 0.975

k’ 1.003 1.005 1.011 1.009
R2 0.999 0.999 0.995 0.995
R2 0 0.960 0.961 0.954 0.962
Rm 0.768 0.774 0.763 0.791

Fig. 17 SHAP analysis; a importance plot and b SHAP influence

predictive models. For instance, large-scale projects would

find the implementation of these models useful in forecasting
the compressive strength of concrete blends, improving effi-
ciency, and reducing project costs. The models could also be
integrated into quality control practices in precast concrete
manufacturing plants, where predicting the strength of differ-
ent concrete mixes is vital for maintaining consistency and
meeting performance specifications. Such predictive tools
could significantly reduce the time and resources required for
extensive laboratory testing and enhance decision-making
processes in material selection. By aligning the predictive
models with specific industry needs and ongoing infras-
tructure projects, this research provides practical, scalable
solutions that are directly applicable to current industry prac-
tices.
Fig. 18 Waterfall plot of SHAP analysis
In practical engineering applications, several challenges
could arise that might not be fully captured in theoreti-
cal model evaluations. All models developed in this study,
SVR-FFA in particular, have shown strong performance in ongoing calibration and validation against newly reported
the tested scenarios, which closely represent practical engi- field data are crucial, allowing the models to adapt to dif-
neering conditions. As a result, these models are expected ferent environmental conditions, material inconsistencies,
to perform effectively in real-world applications. However, and other site-specific factors. Additionally, implementing
in some cases, variability and unforeseen factors can com- a robust monitoring system during the initial deployment
plicate the application of predictive models to practical phases can help identify and correct any discrepancies
problems. In such cases, few strategies can be adopted. First, between predicted and actual performance. Finally, integrat-
ing these models into a broader decision-making framework

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 21 of 28 25

that includes expert judgment, and empirical adjustments can performance may vary when applied to different contexts or
enhance their practical applicability, ensuring that they pro- new datasets not covered in this study. Additionally, while
vide reliable and actionable insights in diverse engineering the SHAP analysis highlighted the importance of key vari-
contexts. ables like the water-to-binder ratio and fly ash-to-binder ratio,
the influence of other potential factors such as curing condi-
tions, temperature, and long-term durability was not fully
5 Conclusion explored. These factors could affect the model’s general-
izability and reliability in practical engineering scenarios.
This study focuses on developing robust and accurate predic- Moreover, the study focused on optimizing the SVR model
tion modeling for the compressive strength (CS) of blended using specific algorithms (GWO, FFA, PSO), which, while
cement concrete (BCC). For model development, this study effective, may not represent the full spectrum of optimization
utilized support vector regression (SVR) as a base model techniques available. Future research could explore alterna-
and optimized with different optimization techniques such tive algorithms or hybrid combinations to further enhance
as firefly algorithm (FFA), particle swarm optimizer (PSO), model performance. Finally, although the developed mod-
and grey wolf optimizer (GWO) to predict the CS of BCC. A els address overfitting concerns, there remains a need for
data-rich dataset containing 1287 data points was collected continuous validation and recalibration as new data becomes
from experimental studies and utilized for model training available. The practical application of these models should
and validation. The dataset was divided into partitions: 70% be approached with caution, and engineers should consider
for training the model and 30% for validating the accuracy conducting supplementary tests to verify model predictions
of the model. 11 diverse input variables were considered for in their specific contexts.
model development of CS of BCC. The model’s performance
Author contributions A.U.K.: conceptualization, data curation, formal
was assessed with various statistical indicators. In addition, analysis; R.A.: methodology, project administration, resources, soft-
SHAP, a model-agonistic approach, was used to interpret the ware, supervision, N.H.: validation, writing—review and editing. M.K:
prediction of the machine learning (ML) model. The study conceptualization, validation, visualization, software, writing—orig-
revealed the hybrid models (SVR-GWO, SVR-FFA) out- inal draft, data curation, formal analysis M.F.J.: formal analysis,
visualization, writing—original draft; N.A.O:. software, validation,
performed the individual SVR model and provided higher visualization; S.S.: data curation, methodology.
correlation and lower error. The mean absolute error (MAE)
value for SVR-GWO and SVR-FFA models remained below Data availability The data will be provided upon request from the cor-
6 MPa for both training and validation while the root mean responding author.
square error (RMSE) value remained below 9 MPa. Similarly,
the correlation value (R) for these two models showed to be Declarations
higher than 0.91 for both sets. Overall, the optimization of
Conflict of interest The authors declare no competing interests.
SVR with GWO and FFA enhanced the accuracy of the indi-
vidual SVR model. In addition, the objective function (OF)
value of the developed models falls below 0.2, indicating
that the issue of model overfitting is addressed. The SHAP
analysis revealed that among the all-input variables, water-to-
cement ratio, fly ash-to-binder ratio, cement-to-binder ratio,
References
and binder content provided higher shape values, indicating Adam AA, Molyneaux TCK, Patnaikuni I, Law DW (2009) Strength,
their higher importance in estimating the CS of BCC. How- sorptivity and carbonation of geopolymer concrete Opportunities
ever, the water-to-binder ratio exhibited a more prominent and Solutions in Structural Engineering and Construction Edi-
influence on the CS of BCC. Finally, the developed models tion 1st Edition First Published ImprintCRC Press Pages 6 eBook
ISBN9780429206528
for the strength of BCC are robust and accurate. The uti- Ahmad A, Farooq F, Niewiadomski P, Ostrowski K, Akbar A, Aslam F,
lization of these models for field application of BCC will Alyousef R (2021a) Prediction of compressive strength of fly ash
optimize both time and resources. based concrete using individual and ensemble algorithm. Materials
The models developed in this study demonstrated superior 14:794. https://doi.org/10.3390/ma14040794
Ahmad W, Ahmad A, Ostrowski KA, Aslam F, Joyklad P, Zajdel P
performance in predicting the CS of BCC, certain limitations (2021b) Application of advanced machine learning approaches
must be acknowledged. Firstly, the model’s predictive accu- to predict the compressive strength of concrete containing sup-
racy is highly dependent on the quality and representativeness plementary cementitious materials. Materials 14:5762. https://doi.
of the dataset used for training and validation. The dataset uti- org/10.3390/ma14195762
Ahmad A, Ahmad W, Aslam F, Joyklad P (2022) Compressive strength
lized in this study is extensive and representative, however, prediction of fly ash-based geopolymer concrete via advanced
variations in some unforeseen conditions may be encoun- machine learning techniques. Case Stud Constr Mater 16:e00840.
tered in real-world applications. Consequently, the model’s https://doi.org/10.1016/j.cscm.2021.e00840

123
 25 Page 22 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Akan MÖA, Dhavale DG, Sarkis J (2017) Greenhouse gas emissions design guidelines for pultruded GFRP RHS profiles subjected to
in the construction industry: an analysis and evaluation of a con- web crippling. Sci Rep 14:10135. https://doi.org/10.1038/s41598-
crete supply chain. J Clean Prod 167:1195–1207. https://doi.org/ 024-59345-4
10.1016/j.jclepro.2017.07.225 Aslam F, Shahab MZ (2024) Supplementary cementitious materials in
Alabduljabbar H, Khan M, Awan HH, Eldin SM, Alyousef R, Mohamed blended cement concrete: advancements in predicting compres-
AM (2023) Predicting ultra-high-performance concrete compres- sive strength through machine learning. Mater Today Commun
sive strength using gene expression programming method. Case 38:107725. https://doi.org/10.1016/j.mtcomm.2023.107725
Stud Constr Mater 18:e02074. https://doi.org/10.1016/j.cscm. Atiş CD (2003) Accelerated carbonation and testing of concrete made
2023.e02074 with fly ash. Constr Build Mater 17:147–152. https://doi.org/10.
Aldrees A, Javed MF, Khan M, Siddiq B (2024) Optimized predic- 1016/S0950-0618(02)00116-2
tion modeling of micropollutant removal efficiency in forward Azareh A, Sardooi ER, Gholami H, Mosavi A, Shahdadi A, Barkhori
osmosis membrane systems using explainable machine learning S (2021) Detection and prediction of lake degradation using land-
algorithms. J Water Process Eng 66:105937. https://doi.org/10. scape metrics and remote sensing dataset. Environ Sci Pollut Res
1016/j.jwpe.2024.105937 28:27283–27298. https://doi.org/10.1007/s11356-021-12522-8
Al-Shamiri AK, Kim JH, Yuan T-F, Yoon YS (2019) Modeling the Baghabra Al-Amoudi OS, Al-Kutti WA, Ahmad S, Maslehuddin
compressive strength of high-strength concrete: an extreme learn- M (2009) Correlation between compressive strength and cer-
ing approach. Constr Build Mater 208:204–219. https://doi.org/ tain durability indices of plain and blended cement concretes.
10.1016/j.conbuildmat.2019.02.165 Cem Concr Compos 31:672–676. https://doi.org/10.1016/j.cemc
Amankwah EO, Bediako M, Kankam CK (2015) Influence of calcined oncomp.2009.05.005
clay pozzolana on strength characteristics of portland cement con- Balakrishnan B, Awal AA (2014) Durability properties of concrete con-
crete. Int J Mater Sci Appl 3:410–419. https://doi.org/10.11648/j. taining high volume Malaysian fly ash. Measurement 2:2–94
ijmsa.20140306.30 Basak D, Pal S, Patranabis DC (2007) Support vector regression. Neural
Amjad SU, Budihardjo MA, Priyambada IB, Asghar R (2022) Synthesis Inf Process Lett Rev 11:203–224
of geopolymer concrete using flyash. IOP Conf Ser Earth Envi- Ben Seghier MEA, Golafshani EM, Jafari-Asl J, Arashpour M (2023)
ron Sci 1098:012072. https://doi.org/10.1088/1755-1315/1098/1/ Metaheuristic-based machine learning modeling of the compres-
012072 sive strength of concrete containing waste glass. Struct Concr
Amran YHM, Alyousef R, Alabduljabbar H, El-Zeadani M (2020) 24:5417–5440. https://doi.org/10.1002/suco.202200260
Clean production and properties of geopolymer concrete; a review. Benson CH (1993) Probability distributions for hydraulic conductivity
J Clean Prod 251:119679. https://doi.org/10.1016/j.jclepro.2019. of compacted soil liners. J Geotech Eng 119:471–486. https://doi.
119679 org/10.1061/(ASCE)0733-9410(1993)119:3(471)
Angulo-Ramirez DE, De Gutiérrez RM, Valencia-Saavedra WG, De Berndt ML (2009) Properties of sustainable concrete contain-
Medeiros MHF, Hoppe-Filho J (2019) Carbonation of hybrid ing fly ash, slag and recycled concrete aggregate. Constr
concrete with high blast furnace slag content and its impact on Build Mater 23:2606–2613. https://doi.org/10.1016/j.conbuild
structural steel corrosion. Mater Constr 69:e182–e182 mat.2009.02.011
Arora VV, Singh B, Patel V (2019) Durability and corrosion studies Bhatt A, Priyadarshini S, Acharath Mohanakrishnan A, Abri A, Sattler
in prestressed concrete made with blended cement. J Asian Concr M, Techapaphawit S (2019) Physical, chemical, and geotechnical
Fed 5:15–24. https://doi.org/10.18702/acf.2019.06.30.15 properties of coal fly ash: a global review. Case Stud Constr Mater
Asghar R, Khattak SU (2023) Predicting the residual flexural capacity 11:e00263. https://doi.org/10.1016/j.cscm.2019.e00263
of fire exposed reinforced concrete beams using gene expression Bilim C, Atiş CD, Tanyildizi H, Karahan O (2009) Predicting the
programming. Department of Civil Engineering, Capital Univer- compressive strength of ground granulated blast furnace slag con-
sity of Science and Technology, Islamabad, p 8 crete using artificial neural network. Adv Eng Softw 40:334–340.
Dong JF, Xu Y, Guan ZW Wang QY (2022) Freeze-thaw behaviour https://doi.org/10.1016/j.advengsoft.2008.05.005
of basalt fibre reinforced recycled aggregate concrete filled CFRP Biswas WK, Alhorr Y, Lawania KK, Sarker PK, Elsarrag E (2017)
tube specimens. Eng Struct 273 Life cycle assessment for environmental product declaration of
Asghar R, Shahzad A, Amjad SU, Akhtar A (2020b) Experimental concrete in the Gulf States. Sustain Cities Soc 35:36–46. https://
determination of the mechanical properties of brick masonry and doi.org/10.1016/j.scs.2017.07.011
low strength concrete block masonry. Department of Civil Engi- Brown SC, Greene JA (2006) The wisdom development scale: translat-
neering, Capital University of Science and Technology, Islamabad, ing the conceptual to the concrete. J Coll Stud Dev 47:1–19
p6 Bucher R, Diederich P, Escadeillas G, Cyr M (2017) Service life of
Asghar R, Javed MF, Alrowais R, Khalil A, Mohamed AM, Mohamed metakaolin-based concrete exposed to carbonation: comparison
A, Vatin NI (2022) Predicting the lateral load carrying capac- with blended cement containing fly ash, blast furnace slag and
ity of reinforced concrete rectangular columns: gene expression limestone filler. Cem Concr Res 99:18–29. https://doi.org/10.1016/
programming. Materials 15:2673. https://doi.org/10.3390/ma1507 j.cemconres.2017.04.013
2673 Burges CJC (1998) A tutorial on support vector machines for pattern
Asghar R, Khan MA, Alyousef R, Javed MF, Ali M (2023) Promoting recognition. Data Min Knowl Disc 2:121–167. https://doi.org/10.
the green construction: scientometric review on the mechani- 1023/A:1009715923555
cal and structural performance of geopolymer concrete. Constr Cain MK, Zhang Z, Yuan K-H (2017) Univariate and multivariate
Build Mater 368:130502. https://doi.org/10.1016/j.conbuildmat. skewness and kurtosis for measuring nonnormality: prevalence,
2023.130502 influence and estimation. Behav Res 49:1716–1735. https://doi.
Asghar R, Javed MF, Saqib M, Alaskar A, Ali M, Nawaz R (2024a) org/10.3758/s13428-016-0814-1
Multi-expression programming based prediction of the seismic Cardinaud G, Rozière E, Martinage O, Loukili A, Barnes-Davin L, Paris
capacity of reinforced concrete rectangular columns. Eng Appl M, Deneele D (2021) Calcined clay—limestone cements: hydra-
Artif Intell 131:107834. https://doi.org/10.1016/j.engappai.2023. tion processes with high and low-grade kaolinite clays. Constr
107834 Build Mater 277:122271. https://doi.org/10.1016/j.conbuildmat.
Asghar R, Javed MF, Ali M, Najeh T, Gamil Y (2024b) Numerical and 2021.122271
artificial intelligence based investigation on the development of

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 23 of 28 25

Celik K, Meral C, Petek Gursel A, Mehta PK, Horvath A, Monteiro PJM Divsholi BS, Lim TYD, Teng S (2014) Durability properties and
(2015) Mechanical properties, durability, and life-cycle assess- microstructure of ground granulated blast furnace slag cement con-
ment of self-consolidating concrete mixtures made with blended crete. Int J Concr Struct Mater 8:157–164. https://doi.org/10.1007/
portland cements containing fly ash and limestone powder. Cem s40069-013-0063-y
Concr Compos 56:59–72. https://doi.org/10.1016/j.cemconcomp. Duan P, Shui Z, Chen W, Shen C (2013) Enhancing microstructure and
2014.11.003 durability of concrete from ground granulated blast furnace slag
Chen L, Kou C-H, Ma S-W (2014) Prediction of slump flow of high- and metakaolin as cement replacement materials. J Market Res
performance concrete via parallel hyper-cubic gene-expression 2:52–59. https://doi.org/10.1016/j.jmrt.2013.03.010
programming. Eng Appl Artif Intell 34:66–74. https://doi.org/10. Durán-Herrera A, Mendoza-Rangel JM, De-Los-Santos EU, Vázquez F,
1016/j.engappai.2014.05.005 Valdez P, Bentz DP (2015) Accelerated and natural carbonation of
Chen L, Wang Z, Khan AA, Khan M, Javed MF, Alaskar A, Eldin concretes with internal curing and shrinkage/viscosity modifiers.
SM (2023) Development of predictive models for sustainable con- Mater Struct 48:1207–1214. https://doi.org/10.1617/s11527-013-
crete via genetic programming-based algorithms. J Market Res 0226-y
24:6391–6410. https://doi.org/10.1016/j.jmrt.2023.04.180 Eguchi K, Takewaka K, Yamaguchi T, Ueda N (2013) A study on
Chithra S, Kumar SRRS, Chinnaraju K, Alfin Ashmita F (2016) A durability of blast furnace slag cement concrete mixed with
comparative study on the compressive strength prediction models metakaolin-based artificial pozzolan in actual marine environment.
for High Performance Concrete containing nano silica and copper In: Third international conference on sustainable construction
slag using regression analysis and artificial neural networks. Constr materials and technologies
Build Mater 114:528–535. https://doi.org/10.1016/j.conbuildmat. Ehteram M, Panahi F, Ahmed AN, Mosavi AH, El-Shafie A (2022)
2016.03.214 Inclusive multiple model using hybrid artificial neural networks
Chou J-S, Tsai C-F, Pham A-D, Lu Y-H (2014) Machine learning in for predicting evaporation. Front Environ Sci. https://doi.org/10.
concrete strength simulations: multi-nation data analytics. Constr 3389/fenvs.2021.789995
Build Mater 73:771–780. https://doi.org/10.1016/j.conbuildmat. Einsfeld RA, Velasco MSL (2006) Fracture parameters for high-
2014.09.054 performance concrete. Cem Concr Res 36:576–583. https://doi.
Collepardi M, Collepardi S, Olagot JO, Simonelli F (2004) The influ- org/10.1016/j.cemconres.2005.09.004
ence of slag and fly ash on the carbonation of concrete. In: Eighth Faleschini F, Zanini MA, Brunelli K, Pellegrino C (2015) Valoriza-
CANMET/ACI international conference on fly ash, silica fume, tion of co-combustion fly ash in concrete production. Mater Des
slag, and natural pozzolans in concrete. Proccedings ACI SP-221- 85:687–694. https://doi.org/10.1016/j.matdes.2015.07.079
29. Las Vegas, Nevada, USA, Citeseer, pp 23–29 Fanghui H, Qiang W, Jingjing F (2015) The differences among the
Crouch LK, Hewitt R, Byard B (2007) High Volume Fly Ash Concrete. roles of ground fly ash in the paste, mortar and concrete. Constr
2007 World of Coal Ash (WOCA), Northern Kentucky, 7–10 May Build Mater 93:172–179. https://doi.org/10.1016/j.conbuildmat.
2007 1–14 2015.05.117
Czarnecki L, Woyciechowski P, Adamczewski G (2018) Risk of con- Farooq F, Ahmed W, Akbar A, Aslam F, Alyousef R (2021) Predic-
crete carbonation with mineral industrial by-products. KSCE J Civ tive modeling for sustainable high-performance concrete from
Eng 22:755–764. https://doi.org/10.1007/s12205-017-1623-5 industrial wastes: a comparison and optimization of models using
de Matos PR, Sakata RD, Prudêncio LR (2019) Eco-efficient low binder ensemble learners. J Clean Prod 292:126032. https://doi.org/10.
high-performance self-compacting concretes. Constr Build Mater 1016/j.jclepro.2021.126032
225:941–955. https://doi.org/10.1016/j.conbuildmat.2019.07.254 Felekoğlu B, Türkel S, Baradan B (2007) Effect of water/cement ratio
Deng P, Chang D-S, Chen X-L, Zhu Z-Y, Asghar R (2024) Analysis on the fresh and hardened properties of self-compacting concrete.
of the dynamic mechanism of square tubular T-joints with chord Build Environ 42:1795–1802. https://doi.org/10.1016/j.buildenv.
flanges subjected to impact loading. Adv Steel Constr 20:21–29. 2006.01.012
https://doi.org/10.18057/IJASC.2024.20.1.3 Fitriani H, Ahmed A, Kolawole O, Hyndman F, Idris Y, Rosidawani
Despotovic M, Nedic V, Despotovic D, Cvetanovic S (2016) Evalua- R (2022) Optimizing compressive strength properties of binary
tion of empirical models for predicting monthly mean horizontal blended cement rice husk concrete for road pavement. Trends Sci
diffuse solar radiation. Renew Sustain Energy Rev 56:246–260. 19:3972–3972. https://doi.org/10.48048/tis.2022.3972
https://doi.org/10.1016/j.rser.2015.11.058 Flower DJM, Sanjayan JG (2007) Green house gas emissions due to
Dhandapani Y, Sakthivel T, Santhanam M, Gettu R, Pillai RG (2018) concrete manufacture. Int J Life Cycle Assess 12:282–288. https://
Mechanical properties and durability performance of concretes doi.org/10.1065/lca2007.05.327
with limestone calcined clay cement (LC3). Cem Concr Res Gandomi AH, Roke DA (2015) Assessment of artificial neural net-
107:136–151. https://doi.org/10.1016/j.cemconres.2018.02.005 work and genetic programming as predictive tools. Adv Eng Softw
Dhanya BS, Santhanam M, Gettu R, Pillai RG (2018) Performance eval- 88:63–72. https://doi.org/10.1016/j.advengsoft.2015.05.007
uation of concretes having different supplementary cementitious Gandomi AH, Alavi AH, Mirzahosseini MR, Nejad FM (2011) Nonlin-
material dosages belonging to different strength ranges. Constr ear genetic-based models for prediction of flow number of asphalt
Build Mater 187:984–995. https://doi.org/10.1016/j.conbuildmat. mixtures. J Mater Civ Eng 23:248–263. https://doi.org/10.1061/
2018.07.185 (ASCE)MT.1943-5533.0000154
Diab AM, Abd Elmoaty AEM, Aly AA (2016) Long term study of Gandomi AH, Babanajad SK, Alavi AH, Farnam Y (2012) Novel
mechanical properties, durability and environmental impact of approach to strength modeling of concrete under triaxial compres-
limestone cement concrete. Alex Eng J 55:1465–1482. https://doi. sion. J Mater Civ Eng 24:1132–1143. https://doi.org/10.1061/(AS
org/10.1016/j.aej.2016.01.031 CE)MT.1943-5533.0000494
Dinakar P, Babu KG, Santhanam M (2007) Corrosion behaviour Garcez MR, Rohden AB, Graupner de Godoy LG (2018) The role of
of blended cements in low and medium strength concretes. concrete compressive strength on the service life and life cycle of
Cem Concr Compos 29:136–145. https://doi.org/10.1016/j.cemc a RC structure: case study. J Clean Prod 172:27–38. https://doi.
oncomp.2006.10.005 org/10.1016/j.jclepro.2017.10.153
Dinakar P, Sethy KP, Sahoo UC (2013) Design of self-compacting García-Segura T, Yepes V, Alcalá J (2014) Life cycle greenhouse gas
concrete with ground granulated blast furnace slag. Mater Des emissions of blended cement concrete including carbonation and
43:161–169. https://doi.org/10.1016/j.matdes.2012.06.049

123
 25 Page 24 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

durability. Int J Life Cycle Assess 19:3–12. https://doi.org/10. Jalal FE, Xu Y, Iqbal M, Jamhiri B, Javed MF (2021a) Predicting the
1007/s11367-013-0614-0 compaction characteristics of expansive soils using two genetic
Gesoğlu M, Güneyisi E, Özbay E (2009) Properties of self-compacting programming-based algorithms. Transport Geotech 30:100608.
concretes made with binary, ternary, and quaternary cementitious https://doi.org/10.1016/j.trgeo.2021.100608
blends of fly ash, blast furnace slag, and silica fume. Con- Jalal FE, Xu Y, Iqbal M, Javed MF, Jamhiri B (2021b) Predictive model-
str Build Mater 23:1847–1854. https://doi.org/10.1016/j.conbuild ing of swell-strength of expansive soils using artificial intelligence
mat.2008.09.015 approaches: ANN, ANFIS and GEP. J Environ Manag 289:112420.
Gettu R, Pillai RG, Santhanam M, Basavaraj AS, Rathnarajan S, https://doi.org/10.1016/j.jenvman.2021.112420
Dhanya BS (2018) Sustainability-based decision support frame- Jau W-C, Fu C-W, Yang C-T (2004) Study of feasibility and mechan-
work for choosing concrete mixture proportions. Mater Struct ical properties for producing high-flowing concrete with recycled
51:165. https://doi.org/10.1617/s11527-018-1291-z coarse aggregates. In: Proceedings of the international workshop
Gholampour A, Ozbakkaloglu T (2017) Performance of sustainable on sustainable development and concrete technology, pp 89–102
concretes containing very high volume class-F fly ash and ground Javed MF, Khan M, Fawad M, Alabduljabbar H, Najeh T, Gamil Y
granulated blast furnace slag. J Clean Prod 162:1407–1417. https:// (2024b) Comparative analysis of various machine learning algo-
doi.org/10.1016/j.jclepro.2017.06.087 rithms to predict strength properties of sustainable green concrete
Golbraikh A, Tropsha A (2002) Beware of q2! J Mol Graph Model containing waste foundry sand. Sci Rep 14:14617. https://doi.org/
20:269–276. https://doi.org/10.1016/S1093-3263(01)00123-1 10.1038/s41598-024-65255-2
Golewski GL (2018) Green concrete composite incorporating fly Javed MF, Khan M, Nehdi ML, Abuhussain M (2024c) Application
ash with high strength and fracture toughness. J Clean Prod of metaheuristic algorithms for compressive strength prediction
172:218–226. https://doi.org/10.1016/j.jclepro.2017.10.065 of steel fiber reinforced concrete exposed to high temperatures.
Güneyisi E, Gesoğlu M, Özbay E (2010) Strength and dry- Mater Today Commun 39:108832. https://doi.org/10.1016/j.mt
ing shrinkage properties of self-compacting concretes incor- comm.2024.108832
porating multi-system blended mineral admixtures. Constr Javed MF, Asghar R, Rehman SKU (2024a) Automating the seis-
Build Mater 24:1878–1887. https://doi.org/10.1016/j.conbuild mic design of reinforced concrete rectangular columns employing
mat.2010.04.015 multi-expression programming: towards the automated design of
Hafez H, Teirelbar A, Kurda R, Tošić N, de la Fuente A (2022) Pre- reinforced concrete structures. In: Advanced optimization appli-
bcc: a novel integrated machine learning framework for predicting cations in engineering. IGI Global, pp 197–236. https://doi.org/10.
mechanical and durability properties of blended cement concrete. 4018/979-8-3693-2161-4.ch009
Constr Build Mater 352:129019. https://doi.org/10.1016/j.conbui Jiang L, Liu Z, Ye Y (2004) Durability of concrete incorporating large
ldmat.2022.129019 volumes of low-quality fly ash. Cem Concr Res 34:1467–1469.
Hameed MM, Abed MA, Al-Ansari N, Alomar MK (2022) Predict- https://doi.org/10.1016/j.cemconres.2003.12.029
ing compressive strength of concrete containing industrial waste Jones MR, Dhir RK, Magee BJ (1997) Concrete containing ternary
materials: novel and hybrid machine learning model. Adv Civ Eng blended binders: resistance to chloride ingress and carbonation.
2022:5586737. https://doi.org/10.1155/2022/5586737 Cem Concr Res 27:825–831. https://doi.org/10.1016/S0008-8846
Harrison TA, Jones MR, Newlands MD, Kandasami S, Khanna G (2012) (97)00075-6
Experience of using the prTS 12390-12 accelerated carbonation Juenger MCG, Snellings R, Bernal SA (2019) Supplementary cemen-
test to assess the relative performance of concrete. Mag Concr Res titious materials: new sources, characterization, and performance
64:737–747. https://doi.org/10.1680/macr.11.00162 insights. Cem Concr Res 122:257–273. https://doi.org/10.1016/j.
Hawileh RA, Abdalla JA, Fardmanesh F, Shahsana P, Khalili A (2017) cemconres.2019.05.008
Performance of reinforced concrete beams cast with different per- Kaewmanee K, Tangtermsirikul S (2014) Properies of binder systems
centages of GGBS replacement to cement. Arch Civ Mech Eng containing cement, fly ash, and limestone powder. Songklanakarin
17:511–519. https://doi.org/10.1016/j.acme.2016.11.006 J Sci Technol 36(5):569–576
Hedayatinia F, Delnavaz M, Emamzadeh SS (2019) Rheological Kaloop MR, Roy B, Chaurasia K, Kim S-M, Jang H-M, Hu J-W,
properties, compressive strength and life cycle assessment of self- Abdelwahed BS (2022) Shear strength estimation of reinforced
compacting concrete containing natural pumice pozzolan. Constr concrete deep beams using a novel hybrid metaheuristic optimized
Build Mater 206:122–129. https://doi.org/10.1016/j.conbuildmat. SVR models. Sustainability 14:5238. https://doi.org/10.3390/su14
2019.02.059 095238
Holt E, Kuosa H, Leivo M, Al-Neshawy F, Piironen J, Sistonen E (2010) Karahan O (2017) Transport properties of high volume fly ash or
Accounting for coupled deterioration mechanisms for durable con- slag concrete exposed to high temperature. Constr Build Mater
crete containing mineral by-products. In: Proceedings of the 2nd 152:898–906. https://doi.org/10.1016/j.conbuildmat.2017.07.051
international conference on sustainable construction materials and Karim R, Islam MdH, Datta SD, Kashem A (2024) Synergistic effects
technologies, Ancona, Italy, pp 1631–1643 of supplementary cementitious materials and compressive strength
Huang Y, Lei Y, Luo X, Fu C (2023) Prediction of compressive strength prediction of concrete using machine learning algorithms with
of rice husk ash concrete: a comparison of different metaheuristic SHAP and PDP analyses. Case Stud Constr Mater 20:e02828.
algorithms for optimizing support vector regression. Case Stud https://doi.org/10.1016/j.cscm.2023.e02828
Constr Mater 18:e02201. https://doi.org/10.1016/j.cscm.2023.e0 Karri SK, Rao GVR, Raju PM (2019) Strength and durability studies
2201 on GGBS concrete. Int J Civ Eng. https://doi.org/10.14445/2348
Hussain K, Choktaweekarn P, Saengsoy W, Srichan T, Tangtermsirikul 8352/IJCE-V2I10P106
S (2013) Effect of cement types, mineral admixtures, and bottom Kavitha OR, Shanthi VM, Arulraj GP, Sivakumar VR (2016)
ash on the curing sensitivity of concrete. Int J Miner Metall Mater Microstructural studies on eco-friendly and durable self-
20:94–105. https://doi.org/10.1007/s12613-013-0699-2 compacting concrete blended with metakaolin. Appl Clay Sci
Inthata S, Kowtanapanich W, Cheerarot R (2013) Prediction of chloride 124–125:143–149. https://doi.org/10.1016/j.clay.2016.02.011
permeability of concretes containing ground pozzolans by artificial Kennedy J, Eberhart R (1995) Particle swarm optimization. In: Proceed-
neural networks. Mater Struct 46:1707–1721. https://doi.org/10. ings of ICNN’95—international conference on neural networks,
1617/s11527-012-0009-x vol 4, pp 1942–1948. https://doi.org/10.1109/ICNN.1995.488968

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 25 of 28 25

Khan MA, Farooq F, Javed MF, Zafar A, Ostrowski KA, Aslam F, Ling W, Pei T, Yan Y (2004) Application of ground granulated
Malazdrewicz S, Maślak M (2022) Simulation of depth of wear blast furnace slag in high-performance concrete in China. In:
of eco-friendly concrete using machine learning based computa- International workshop on sustainable development and concrete
tional approaches. Materials 15:58. https://doi.org/10.3390/ma15 technology, organized by China Building Materials Academy,
010058 PRC, pp 309–317
Khodabakhshian A, de Brito J, Ghalehnovi M, Asadi Shamsabadi E Liu S, Wang Z, Li X (2014) Long-term properties of concrete containing
(2018) Mechanical, environmental and economic performance of ground granulated blast furnace slag and steel slag. Mag Concr Res
structural concrete containing silica fume and marble industry 66:1095–1103. https://doi.org/10.1680/macr.14.00074
waste powder. Constr Build Mater 169:237–251. https://doi.org/ Löfgren I, Esping O, Lindvall A (2016) The influence of carbonation
10.1016/j.conbuildmat.2018.02.192 and age on salt frost scaling of concrete with mineral additions.
Khodair Y, Bommareddy B (2017) Self-consolidating concrete using Mater Syst Struct Civ Eng 2016:91–100
recycled concrete aggregate and high volume of fly ash, and slag. Long G, Gao Y, Xie Y (2015) Designing more sustainable and greener
Constr Build Mater 153:307–316. https://doi.org/10.1016/j.conb self-compacting concrete. Constr Build Mater 84:301–306. https://
uildmat.2017.07.063 doi.org/10.1016/j.conbuildmat.2015.02.072
Khunthongkeaw J, Tangtermsirikul S, Leelawat T (2006) A study Long W-J, Khayat KH, Yahia A, Xing F (2017) Rheological approach
on carbonation depth prediction for fly ash concrete. Constr in proportioning and evaluating prestressed self-consolidating con-
Build Mater 20:744–753. https://doi.org/10.1016/j.conbuildmat. crete. Cem Concr Compos 82:105–116. https://doi.org/10.1016/j.
2005.01.052 cemconcomp.2017.05.008
Kou S-C, Poon C-S (2013) Long-term mechanical and durability Lothenbach B, Scrivener K, Hooton RD (2011) Supplementary cemen-
properties of recycled aggregate concrete prepared with the incor- titious materials. Cem Concr Res 41:1244–1256. https://doi.org/
poration of fly ash. Cement Concr Compos 37:12–19. https://doi. 10.1016/j.cemconres.2010.12.001
org/10.1016/j.cemconcomp.2012.12.011 Lübeck A, Gastaldini ALG, Barin DS, Siqueira HC (2012) Compressive
Kou SC, Poon CS, Chan D (2007) Influence of fly ash as cement replace- strength and electrical properties of concrete with white Portland
ment on the properties of recycled aggregate concrete. J Mater cement and blast-furnace slag. Cem Concr Compos 34:392–399.
Civ Eng 19:709–717. https://doi.org/10.1061/(ASCE)0899-1561 https://doi.org/10.1016/j.cemconcomp.2011.11.017
(2007)19:9(709) Lundberg S (2017) A unified approach to interpreting model predic-
Kou S, Poon C, Agrela F (2011) Comparisons of natural and recycled tions. arXiv Preprint arXiv:1705.07874
aggregate concretes prepared with the addition of different mineral Marinković S, Dragaš J, Ignjatović I, Tošić N (2017) Environmental
admixtures. Cem Concr Compos 33:788–795. https://doi.org/10. assessment of green concretes for structural use. J Clean Prod
1016/j.cemconcomp.2011.05.009 154:633–649. https://doi.org/10.1016/j.jclepro.2017.04.015
Kumar S, Rai B, Biswas R, Samui P, Kim D (2020) Prediction of rapid Marques PF, Chastre C, Nunes  (2013) Carbonation service life mod-
chloride permeability of self-compacting concrete using multivari- elling of RC structures for concrete with Portland and blended
ate adaptive regression spline and minimax probability machine cements. Cem Concr Compos 37:171–184. https://doi.org/10.
regression. J Build Eng 32:101490. https://doi.org/10.1016/j.jobe. 1016/j.cemconcomp.2012.10.007
2020.101490 Mazloom M, Ramezanianpour AA, Brooks JJ (2004) Effect of silica
Tipu RK, Batra V, Suman, Pandya KS, Panchal VR (2023c) Shear capac- fume on mechanical properties of high-strength concrete. Cement
ity prediction for FRCM-strengthened RC beams using hybrid Concr Compos 26:347–357. https://doi.org/10.1016/S0958-9465
ReLU-activated BPNN model. Structures 58:105432. https://doi. (03)00017-9
org/10.1016/j.istruc.2023.105432 McCarthy MJ, Dhir RK (2005) Development of high volume fly ash
Kurda R, Silvestre JD, de Brito J (2018) Life cycle assessment of con- cements for use in concrete construction. Fuel 84:1423–1432.
crete made with high volume of recycled concrete aggregates and https://doi.org/10.1016/j.fuel.2004.08.029
fly ash. Resour Conserv Recycl 139:407–417. https://doi.org/10. McNally C, Sheils E (2012) Probability-based assessment of the dura-
1016/j.resconrec.2018.07.004 bility characteristics of concretes manufactured using CEM II and
Kurda R, de Brito J, Silvestre JD (2019) CONCRETop—a multi-criteria GGBS binders. Constr Build Mater 30:22–29. https://doi.org/10.
decision method for concrete optimization. Environ Impact Assess 1016/j.conbuildmat.2011.11.029
Rev 74:73–85. https://doi.org/10.1016/j.eiar.2018.10.006 Meddah MS, Lmbachiya MC, Dhir RK (2014) Potential use of binary
Lee S, Park W, Lee H (2013) Life cycle CO2 assessment method for and composite limestone cements in concrete production. Constr
concrete using CO2 balance and suggestion to decrease LCCO2 Build Mater 58:193–205. https://doi.org/10.1016/j.conbuildmat.
of concrete in South-Korean apartment. Energy Build 58:93–102. 2013.12.012
https://doi.org/10.1016/j.enbuild.2012.11.034 Megat Johari MA, Brooks JJ, Kabir S, Rivard P (2011) Influence of
Leemann A, Nygaard P, Kaufmann J, Loser R (2015) Relation between supplementary cementitious materials on engineering properties of
carbonation resistance, mix design and exposure of mortar and high strength concrete. Constr Build Mater 25:2639–2648. https://
concrete. Cem Concr Compos 62:33–43. https://doi.org/10.1016/ doi.org/10.1016/j.conbuildmat.2010.12.013
j.cemconcomp.2015.04.020 Miller SA (2018) Supplementary cementitious materials to mitigate
Leung HY, Kim J, Nadeem A, Jaganathan J, Anwar MP (2016) Sorptiv- greenhouse gas emissions from concrete: can there be too much of
ity of self-compacting concrete containing fly ash and silica fume. a good thing? J Clean Prod 178:587–598. https://doi.org/10.1016/
Constr Build Mater 113:369–375. https://doi.org/10.1016/j.conb j.jclepro.2018.01.008
uildmat.2016.03.071 Miller SA, John VM, Pacca SA, Horvath A (2018) Carbon diox-
Lima C, Caggiano A, Faella C, Martinelli E, Pepe M, Realfonzo R ide reduction potential in the global cement industry by 2050.
(2013) Physical properties and mechanical behaviour of concrete Cem Concr Res 114:115–124. https://doi.org/10.1016/j.cemcon
made with recycled aggregates and fly ash. Constr Build Mater res.2017.08.026
47:547–559. https://doi.org/10.1016/j.conbuildmat.2013.04.051 Mirjalili S, Mirjalili SM, Lewis A (2014) Grey wolf optimizer. Adv
Limbachiya M, Meddah MS, Ouchagour Y (2012) Use of recy- Eng Softw 69:46–61. https://doi.org/10.1016/j.advengsoft.2013.
cled concrete aggregate in fly-ash concrete. Constr Build Mater 12.007
27:439–449. https://doi.org/10.1016/j.conbuildmat.2011.07.023 Mittal A, Kaisare MB, Shetti R (2005) Experimental study on use of
fly ash in concrete, Tarapur Atomic Power Project 3

123
 25 Page 26 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Moffatt EG, Thomas MDA, Fahim A (2017) Performance of high- Energy Rev 16:2940–2946. https://doi.org/10.1016/j.rser.2012.02
volume fly ash concrete in marine environment. Cem Concr Res .014
102:127–135. https://doi.org/10.1016/j.cemconres.2017.09.008 Parron-Rubio ME, Perez-Garcia F, Gonzalez-Herrera A, Oliveira MJ,
Mohamed OA, Ati M, Hawat WA (2018) Implementation of artificial Rubio-Cintas MD (2019) Slag substitution as a cementing material
neural networks for prediction of chloride penetration in concrete. in concrete: mechanical, physical and environmental properties.
Int J Eng Technol 7:47–52. https://doi.org/10.14419/ijet.v7i2.28. Materials 12:2845. https://doi.org/10.3390/ma12182845
12880 Patil YO, Patil PN, Dwivedi AK (2013) GGBS as partial replacement
Mohammadi J, South W (2017) Life cycle assessment (LCA) of bench- of OPC in cement concrete—an experimental study. Int J Sci Res
mark concrete products in Australia. Int J Life Cycle Assess 2:189–191
22:1588–1608. https://doi.org/10.1007/s11367-017-1266-2 Pillai RG, Gettu R, Santhanam M, Rengaraju S, Dhandapani Y,
Murad Y, Imam R, Abu Hajar H, Habeh D, Hammad A, Shawash Rathnarajan S, Basavaraj AS (2019) Service life and life cycle
Z (2019) Predictive compressive strength models for green con- assessment of reinforced concrete systems with limestone calcined
crete. Int J Struct Integr 11:169–184. https://doi.org/10.1108/IJSI- clay cement (LC3). Cem Concr Res 118:111–119. https://doi.org/
05-2019-0044 10.1016/j.cemconres.2018.11.019
Nafees A, Javed MF, Khan S, Nazir K, Farooq F, Aslam F, Musarat Poon CS, Kou SC, Lam L (2006) Compressive strength, chloride dif-
MA, Vatin NI (2021) Predictive modeling of mechanical prop- fusivity and pore structure of high performance metakaolin and
erties of silica fume-based green concrete using artificial intelli- silica fume concrete. Constr Build Mater 20:858–865. https://doi.
gence approaches: MLPNN, ANFIS, and GEP. Materials 14:7531. org/10.1016/j.conbuildmat.2005.07.001
https://doi.org/10.3390/ma14247531 Preez D, Phillip M (2019) Sensitivity of strength and durability prop-
Navarro IJ, Yepes V, Martí JV (2018) Life cycle cost assessment of pre- erties of blended cement concrete to changes in water/binder ratio
ventive strategies applied to prestressed concrete bridges exposed and binder content. https://hdl.handle.net/10539/28070. Accessed
to chlorides. Sustainability 10:845. https://doi.org/10.3390/su10 3 Sept 2024
030845 Quan H, Kasami H (2014) Experimental study on durability improve-
Nepomuceno MCS, Pereira-de-Oliveira LA, Lopes SMR (2014) ment of fly ash concrete with durability improving admixture. Sci
Methodology for the mix design of self-compacting concrete using World J 2014:818103. https://doi.org/10.1155/2014/818103
different mineral additions in binary blends of powders. Con- Rathnarajan S, Vaddey NP, Pillai RG, Gettu R, Santhanam M (2017)
str Build Mater 64:82–94. https://doi.org/10.1016/j.conbuildmat. Modelling carbonation rates in concretes with similar strength and
2014.04.021 with and without slag. In: Conference: ICACMS 2017, Chennai,
Nochaiya T, Wongkeo W, Chaipanich A (2010) Utilization of fly ash India
with silica fume and properties of Portland cement–fly ash–silica Roy PP, Roy K (2008) On some aspects of variable selection for partial
fume concrete. Fuel 89:768–774. https://doi.org/10.1016/j.fuel. least squares regression models. QSAR Comb Sci 27:302–313.
2009.10.003 https://doi.org/10.1002/qsar.200710043
Omran BA, Chen Q, Jin R (2016) Comparison of data mining techniques Rozière E, Loukili A, Cussigh F (2009) A performance based approach
for predicting compressive strength of environmentally friendly for durability of concrete exposed to carbonation. Constr Build
concrete. J Comput Civ Eng 30:04016029. https://doi.org/10.1061/ Mater 23:190–199. https://doi.org/10.1016/j.conbuildmat.2008.
(ASCE)CP.1943-5487.0000596 01.006
Oner A, Akyuz S (2007) An experimental study on optimum usage of Ruixia H (2010) A study on carbonation for low calcium fly ash con-
GGBS for the compressive strength of concrete. Cem Concr Com- crete under different temperature and relative humidity. Electron J
pos 29:505–514. https://doi.org/10.1016/j.cemconcomp.2007.01 Geotech Eng (EJGE) 15:1871–1877
.001 Saha P, Debnath P, Thomas P (2020) Prediction of fresh and hard-
Oner A, Akyuz S, Yildiz R (2005) An experimental study on strength ened properties of self-compacting concrete using support vector
development of concrete containing fly ash and optimum usage of regression approach. Neural Comput Appl 32:7995–8010. https://
fly ash in concrete. Cem Concr Res 35:1165–1171. https://doi.org/ doi.org/10.1007/s00521-019-04267-w
10.1016/j.cemconres.2004.09.031 Şahmaran M, Yaman İÖ, Tokyay M (2009) Transport and mechanical
Özbay E, Erdemir M, Durmuş Hİ (2016) Utilization and efficiency of properties of self consolidating concrete with high volume fly ash.
ground granulated blast furnace slag on concrete properties—a Cement Concr Compos 31:99–106. https://doi.org/10.1016/j.cemc
review. Constr Build Mater 105:423–434. https://doi.org/10.1016/ oncomp.2008.12.003
j.conbuildmat.2015.12.153 Dong JF, Wang QY, Guan ZW, Chai HK (2022) High-temperature
Ozcan G, Kocak Y, Gulbandilar E (2017) Estimation of compressive behaviour of basalt fibre reinforced concrete made with recycled
strength of BFS and WTRP blended cement mortars with machine aggregates from earthquake waste. J Build Eng 48
learning models. Comput Concr 19:275–282. https://doi.org/10.12 Samad S, Shah A, Limbachiya MC (2017) Strength development char-
989/cac.2017.19.3.275 acteristics of concrete produced with blended cement using ground
Pal SC, Mukherjee A, Pathak SR (2003) Investigation of hydraulic granulated blast furnace slag (GGBS) under various curing con-
activity of ground granulated blast furnace slag in concrete. Cem ditions. Sādhanā 42:1203–1213. https://doi.org/10.1007/s12046-
Concr Res 33:1481–1486. https://doi.org/10.1016/S0008-8846(03 017-0667-z
)00062-0 San Nicolas R, Cyr M, Escadeillas G (2014) Performance-
Panesar DK, Zhang R (2020) Performance comparison of cement based approach to durability of concrete containing flash-
replacing materials in concrete: Limestone fillers and supple- calcined metakaolin as cement replacement. Constr Build Mater
mentary cementing materials—a review. Constr Build Mater 55:313–322. https://doi.org/10.1016/j.conbuildmat.2014.01.063
251:118866. https://doi.org/10.1016/j.conbuildmat.2020.118866 Sathyan D, Anand KB (2019) Influence of superplasticizer family on the
Panesar DK, Seto KE, Churchill CJ (2017) Impact of the selection of durability characteristics of fly ash incorporated cement concrete.
functional unit on the life cycle assessment of green concrete. Int J Constr Build Mater 204:864–874. https://doi.org/10.1016/j.conb
Life Cycle Assess 22:1969–1986. https://doi.org/10.1007/s11367- uildmat.2019.01.171
017-1284-0
Park J, Tae S, Kim T (2012) Life cycle CO2 assessment of concrete by
compressive strength on construction site in Korea. Renew Sustain

123
Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25 Page 27 of 28 25

Scrivener KL, John VM, Gartner EM (2018) Eco-efficient cements: Tipu RK, Panchal VR, Pandya KS (2024a) Enhancing chloride concen-
potential economically viable solutions for a low-CO2 cement- tration prediction in marine concrete using conjugate gradient-
based materials industry. Cem Concr Res 114:2–26. https://doi. optimized backpropagation neural network. Asian J Civ Eng
org/10.1016/j.cemconres.2018.03.015 25:637–656. https://doi.org/10.1007/s42107-023-00801-3
Shaikh FUA, Supit SWM (2015) Compressive strength and durability Tipu RK, Batra V, Suman, Pandya KS, Panchal VR (2023a) Efficient
properties of high volume fly ash (HVFA) concretes containing compressive strength prediction of concrete incorporating recycled
ultrafine fly ash (UFFA). Constr Build Mater 82:192–205. https:// coarse aggregate using Newton’s boosted backpropagation neu-
doi.org/10.1016/j.conbuildmat.2015.02.068 ral network (NB-BPNN). Structures 58:105559. https://doi.org/
Shi H, Xu B, Zhou X (2009) Influence of mineral admixtures on 10.1016/j.istruc.2023.105559
compressive strength, gas permeability and carbonation of high Tipu RK, Batra V, Suman, Panchal VR, Pandya KS (2024b) Predictive
performance concrete. Constr Build Mater 23:1980–1985. https:// modelling of surface chloride concentration in marine concrete
doi.org/10.1016/j.conbuildmat.2008.08.021 structures: a comparative analysis of machine learning approaches.
Siddique R (2004) Performance characteristics of high-volume class F Asian J Civ Eng 25:1443–1465. https://doi.org/10.1007/s42107-
fly ash concrete. Cem Concr Res 34:487–493. https://doi.org/10. 023-00854-4
1016/j.cemconres.2003.09.002 Turner LK, Collins FG (2013) Carbon dioxide equivalent (CO2-e)
Siddique R, Aggarwal P, Aggarwal Y (2011) Prediction of compressive emissions: a comparison between geopolymer and OPC cement
strength of self-compacting concrete containing bottom ash using concrete. Constr Build Mater 43:125–130. https://doi.org/10.1016/
artificial neural networks. Adv Eng Softw 42:780–786. https://doi. j.conbuildmat.2013.01.023
org/10.1016/j.advengsoft.2011.05.016 Uysal M, Sumer M (2011) Performance of self-compacting
Silva MG, Saade MRM, Gomes V (2013) Influence of service life, concrete containing different mineral admixtures. Constr
strength and cement type on life cycle environmental performance Build Mater 25:4112–4120. https://doi.org/10.1016/j.conbuild
of concrete. Rev IBRACON Estrut Mater 6:844–853. https://doi. mat.2011.04.032
org/10.1590/S1983-41952013000600002 Van den Heede P, De Keersmaecker M, Elia A, Adriaens A, De
Smola AJ, Schölkopf B (2004) A tutorial on support vector regression. Belie N (2017) Service life and global warming potential
Stat Comput 14:199–222. https://doi.org/10.1023/B:STCO.0000 of chloride exposed concrete with high volumes of fly ash.
035301.49549.88 Cement Concr Compos 80:210–223. https://doi.org/10.1016/j.ce
Sonebi et al. (2008): 11DBMC International Conference on Durability mconcomp.2017.03.020
of Building Materials and Components ISTANBUL, Turkey 11–14 Van den Heede P, De Schepper M, De Belie N (2019) Accelerated
May 2008 and natural carbonation of concrete with high volumes of fly ash:
Song H, Ahmad A, Farooq F, Ostrowski KA, Maślak M, Czarnecki chemical, mineralogical and microstructural effects. R Soc Open
S, Aslam F (2021) Predicting the compressive strength of con- Sci 6:181665. https://doi.org/10.1098/rsos.181665
crete with fly ash admixture using machine learning algorithms. Vejmelková E, Pavlíková M, Keršner Z, Rovnaníková P, Ondráček M,
Constr Build Mater 308:125021. https://doi.org/10.1016/j.conbui Sedlmajer M, Černý R (2009) High performance concrete con-
ldmat.2021.125021 taining lower slag amount: a complex view of mechanical and
Soutsos M, Kanavaris F, Hatzitheodorou A (2018) Critical analysis durability properties. Constr Build Mater 23:2237–2245. https://
of strength estimates from maturity functions. Case Stud Constr doi.org/10.1016/j.conbuildmat.2008.11.018
Mater 9:e00183. https://doi.org/10.1016/j.cscm.2018.e00183 Vejmelková E, Keppert M, Grzeszczyk S, Skaliński B, Černý R
Sugi H, Tsukagoshi M, Ueda T (2013) Durability of concrete compos- (2011) Properties of self-compacting concrete mixtures con-
ites containing fly ash and blast furnace slag for use in for precast taining metakaolin and blast furnace slag. Constr Build Mater
concrete products. In: Proceedings of 3rd international confer- 25:1325–1331. https://doi.org/10.1016/j.conbuildmat.2010.09
ence on sustainable construction materials and technology, Kyoto, .012
Japan Vivek SS, Dhinakaran G (2017) Durability characteristics of binary
Sui S, Georget F, Maraghechi H, Sun W, Scrivener K (2019) Towards blend high strength SCC. Constr Build Mater 146:1–8. https://doi.
a generic approach to durability: Factors affecting chloride trans- org/10.1016/j.conbuildmat.2017.04.063
port in binary and ternary cementitious materials. Cem Concr Res Vollpracht A, Soutsos M, Kanavaris F (2018) Strength development
124:105783. https://doi.org/10.1016/j.cemconres.2019.105783 of GGBS and fly ash concretes and applicability of fib model
Sujjavanich S, Suwanvitaya P, Chaysuwan D, Heness G (2017) Syner- code’s maturity function—a critical review. Constr Build Mater
gistic effect of metakaolin and fly ash on properties of concrete. 162:830–846. https://doi.org/10.1016/j.conbuildmat.2017.12.054
Constr Build Mater 155:830–837. https://doi.org/10.1016/j.conb Vu DD, Stroeven P, Bui VB (2001) Strength and durability aspects
uildmat.2017.08.072 of calcined kaolin-blended Portland cement mortar and concrete.
Tae S, Baek C, Shin S (2011) Life cycle CO2 evaluation on rein- Cement Concr Compos 23:471–478. https://doi.org/10.1016/S0
forced concrete structures with high-strength concrete. Environ 958-9465(00)00091-3
Impact Assess Rev 31:253–260. https://doi.org/10.1016/j.eiar. Wang X-Y (2019) Simulation for optimal mixture design of low-CO2
2010.07.002 high-volume fly ash concrete considering climate change and CO2
Teng S, Lim TYD, Sabet Divsholi B (2013) Durability and mechanical uptake. Cement Concr Compos 104:103408. https://doi.org/10.
properties of high strength concrete incorporating ultra fine ground 1016/j.cemconcomp.2019.103408
granulated blast-furnace slag. Constr Build Mater 40:875–881. Wongkeo W, Thongsanitgarn P, Ngamjarurojana A, Chaipanich A
https://doi.org/10.1016/j.conbuildmat.2012.11.052 (2014) Compressive strength and chloride resistance of self-
Thomas A, Lombardi DR, Hunt D, Gaterell M (2009) Estimating carbon compacting concrete containing high level fly ash and silica fume.
dioxide emissions for aggregate use. Proc Inst Civ Eng Eng Sustain Mater Des 64:261–269. https://doi.org/10.1016/j.matdes.2014.07
162:135–144. https://doi.org/10.1680/ensu.2009.162.3.135 .042
Tipu RK, Panchal VR, Pandya KS (2023b) Multi-objective optimized Wu Y, Zhou Y (2022) Hybrid machine learning model and Shapley
high-strength concrete mix design using a hybrid machine learning additive explanations for compressive strength of sustainable con-
and metaheuristic algorithm. Asian J Civ Eng 24:849–867. https:// crete. Constr Build Mater 330:127298. https://doi.org/10.1016/j.
doi.org/10.1007/s42107-022-00535-8 conbuildmat.2022.127298

123
 25 Page 28 of 28 Multiscale and Multidisciplinary Modeling, Experiments and Design (2025) 8:25

Wu Y, Zhou Y (2023) Prediction and feature analysis of punching shear Yoo S-W, Ryu G-S, Choo JF (2015) Evaluation of the effects of
strength of two-way reinforced concrete slabs using optimized high-volume fly ash on the flexural behavior of reinforced con-
machine learning algorithm and Shapley additive explanations. crete beams. Constr Build Mater 93:1132–1144. https://doi.org/
Mech Adv Mater Struct 30:3086–3096. https://doi.org/10.1080/ 10.1016/j.conbuildmat.2015.05.021
15376494.2022.2068209 Zhang Y, Song C, Zhang J, Bo H, Ji LL (2018) Exercise training induced
Wu K-R, Chen B, Yao W, Zhang D (2001) Effect of coarse aggregate anti-inflammatory IL-6 in aged skeletal muscle: role of mitochon-
type on mechanical properties of high-performance concrete. Cem drial homeostasis: 872 Board #133 May 30 2:00 PM–3:30 PM.
Concr Res 31:1421–1425. https://doi.org/10.1016/S0008-8846(01 Med Sci Sports Exerc 50:198. https://doi.org/10.1249/01.mss.00
)00588-9 00535737.45954.59
Yang X-S (2009) Firefly algorithms for multimodal optimization. In: Zhao H, Sun W, Wu X, Gao B (2015) The properties of the self-
Watanabe O, Zeugmann T (eds) Stochastic algorithms: founda- compacting concrete with fly ash and ground granulated blast
tions and applications. Springer, Berlin, pp 169–178. https://doi. furnace slag mineral admixtures. J Clean Prod 95:66–74. https://
org/10.1007/978-3-642-04944-6_14 doi.org/10.1016/j.jclepro.2015.02.050
Yang X-S, Karamanoglu M (2013) 1—Swarm intelligence and bio-
inspired computation: an overview. In: Yang X-S, Cui Z, Xiao R,
Gandomi AH, Karamanoglu M (eds) Swarm intelligence and bio- Publisher’s Note Springer Nature remains neutral with regard to juris-
inspired computation. Elsevier, Oxford, pp 3–23. https://doi.org/ dictional claims in published maps and institutional affiliations.
10.1016/B978-0-12-405163-8.00001-6
Yazıcı H (2008) The effect of silica fume and high-volume Class Springer Nature or its licensor (e.g. a society or other partner) holds
C fly ash on mechanical properties, chloride penetration and exclusive rights to this article under a publishing agreement with the
freeze–thaw resistance of self-compacting concrete. Constr Build author(s) or other rightsholder(s); author self-archiving of the accepted
Mater 22:456–462. https://doi.org/10.1016/j.conbuildmat.2007. manuscript version of this article is solely governed by the terms of such
01.002 publishing agreement and applicable law.
Yeau KY, Kim EK (2005) An experimental study on corrosion resistance
of concrete with ground granulate blast-furnace slag. Cem Concr
Res 35:1391–1399. https://doi.org/10.1016/j.cemconres.2004.11
.010
Yesilyurt MK (2018) The evaluation of a direct injection diesel engine
operating with waste cooking oil biodiesel in point of the envi-
ronmental and enviroeconomic aspects. Energy Sources Part A
Recovery Util Environ Effects 40:654–661. https://doi.org/10.
1080/15567036.2018.1454546

123