International Journal on Interactive Design and Manufacturing (IJIDeM)

https://doi.org/10.1007/s12008-023-01386-6

ORIGINAL PAPER

Development of performance-based models for green concrete using

multiple linear regression and artificial neural network
Priyanka Singh1 · Abiola Adebanjo2,3 · Nasir Shafiq2 · Siti Nooriza Abd Razak2 · Vicky Kumar2 ·
Syed Ahmad Farhan2 · Ifeoluwa Adebanjo3 · Archisha Singh4 · Saurav Dixit5,6 · Subhav Singh7 ·
Meshcheryakova Tatyana Sergeevna8

Received: 16 March 2023 / Accepted: 25 May 2023

© The Author(s), under exclusive licence to Springer-Verlag France SAS, part of Springer Nature 2023

Abstract
The impact of process inputs and critical performance parameters on product quality is an important aspect of production and
this is also true for concrete. There has been an increasing emphasis on the use of machine learning algorithms for modelling in
order to improve production quality and processes. Multiple linear regression and artificial neural network are used as predictive
models in this study to generalise the relationship between seven process variables and three performance parameters in green
concrete production. Models were developed by using 103 experimental datasets obtained from the production of green
concrete. Indices such as p value, residual predicted plots, R-squared and mean squared error were used to evaluate the
models. Due to the masking effect and non-linear nature of the rheologic properties, multiple linear regression was ineffective
at predicting the rheologic behaviour of green concrete, as evidenced by low R2 values of 0.323 and 0.506 obtained for
slump and flow properties, respectively. However, the model was significant at predicting the compressive strength with an
R2 value of 0.898. Conversely, artificial neural network models with varying amount of hidden layer neurons generalized the
relationship between the process variables and performance parameters much better. Optimal network architecture of 7-4-1,
7-2-1 and 7-3-1 with corresponding R2 values of 0.918, 0.826 and 0.945 were obtained for slump, flow and compressive
strength, respectively. Therefore, in developing performance-based models to produce green concrete the use of ANN is
considered a better alternative particularly when there are limited number of process inputs.

Keywords Artificial neural network · Compressive strength · Green concrete · Rheological properties · Multiple linear
regression · Performance-based model

Abbreviations

ANN Artificial neural network

B Saurav Dixit CPPs Critical process parameters
[email protected] GGBS Ground granulated blast-furnace slag
Subhav Singh MLR Multiple linear regression
[email protected] MSE Mean-squared error
Meshcheryakova Tatyana Sergeevna
[email protected]
5 Department of Industrial and Systems Engineering, Khalifa
1 Department of Civil Engineering, Amity University, Noida, University, P.O. Box 127788, Abu Dhabi, United Arab
India Emirates
2 6 Division of Research and Innovation, Uttaranchal University,
Department of Civil and Environmental Engineering,
Universiti Teknologi PETRONAS, Seri Iskandar, Malaysia Dehradun, India
3 7 Lovely Professional University, Ludhiyana, India
Department of Civil Engineering, Osun State University,
Osogbo, Nigeria 8 National Research University, Moscow State University of
4 Department of Computer Science and Engineering, Indian Civil Engineering, Yaroslavskoe shosse, 26, Moscow 129337,
Institute of Technology, Kanpur, India Russia

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

ONA Optimal network architecture of cement alternatives on specific properties of concrete

R2 Coefficient of determination [16–20]. The possibility of creating performance-based mod-
els to monitor CPPs that influence concrete quality during
batch production is attainable through the utilisation of
machine learning algorithms. For instance, the mechanical
properties of alkali-activated mortars containing FA, GGBS,
1 Introduction and POFA have been modelled to a high degree of accuracy
by using ANN and the Krill Herd Algorithm (KHA). Results
The production of concrete and its composites contributes revealed that the hybrid of the machine learning models of
more than 10% of the carbon dioxide (CO2 ) hence causing ANN and KHA produced satisfactory results largely because
devasting environmental and ecological effects on the climate of the weight and bias obtained from the models [21].
globally [1, 2]. Considering the environmental and ecological Similarly, compressive strength of light-weight alkali
concerns inherent in concrete production. The implementa- activated concrete produced by replacing conventional aggre-
tion of industrial ecology, green chemistry, energy reduction gates with sawdust waste has also been predicted using ANN.
and nanoengineering are some of the technologies that have Linear coefficient of determination (R2 ) of 0.99 and 0.98
been explored to develop alternatives that can reduce the were obtained for training and validation phases, respec-
environmental impacts that have plagued the use of cement tively. An adaptive neuro-fuzzy inference system (ANFIS)
in concrete production [2, 3]. By adopting these technolo- was used in predicting the compressive strength of concrete
gies, green cementitious composites which are obtained as with agriculture-based supplementary cementitious materi-
residues from natural, industrial and agricultural processes als (SCMs). The performance of the developed ANFIS model
have produced a wide range of alternative materials, such as was evaluated using R2 , root mean square error (RMSE),
metakaolin (MK), fly ash (FA), rice husk ash (RHA), palm mean absolute error (MAE) and mean square error (MSE).
oil fuel ash (POFA), bagasse ash (BA), sludge waste ash Results showed that with R2 of 0.988, RMSE of 0.63, MAE
(SWA), ground granulated blast-furnace slag (GGBS), vol- of 0.22 and MSE of 0.39, the model was able to generalize
canic ash (VA), ceramic waste powder (CWP), silica fume the relationship between the input and the output parameters
(SF) and wood waste (WWA) as potential alternatives for accurately [18]. Sadowski et al. [20] proposed the use ultra-
cement [4–10]. sonic pulse velocity (UPV) and ANN has effective tools in
The utilisation of cement alternatives, either partially determining the compressive strength of eco-friendly mortar
or completely, has resulted in the production of greener samples with high volumes of waste quartz mineral dust and
concretes with improved flowability, strength, stiffness and reported high R2 of 0.93, 0.91 and 0.94 for training, valida-
sustainability [2, 11–14]. Hossain and Lachemi [15] reported tion and testing, respectively. According to the authors, the
that the use of VA in concrete improved the rheological as high values of R2 suggest that the technique adopted is reli-
well as mechanical properties due to the improvement of able for modelling the compressive strength of the concrete.
the pore structure. Elsewhere, the use of FA and GGBS in Amar et al. [17] compiled a comprehensive list of the
alkali-activated concrete has also been attributed to produc- compressive strengths of concrete produced using admix-
ing concrete with improved properties, such as faster setting tures, such as FA, MK, GGBS, SF, and CWP, as well as
time, increased strength, low permeability and higher resis- limestone filler and recycled aggregates, based on secondary
tance to fire [10]. The production of green concrete has data sources. Following the collection of these data, an ANN
predominantly depended on the knowledge and expertise was developed in order to make a prediction regarding the
of individuals. However, with the advent of contemporary compressive strength of concrete made from waste materi-
technologies, such as machine learning, predictive modelling als. According to the findings, ANN is effective at predicting
and expedited data processing. The quality, performance and the compressive strength of waste-based concrete. The accu-
production process for green concrete can be improved by racy of the developed model is impressive considering the
adopting models that proactively optimise processes prior to R2 value of 0.9888 that was acquired from the training. The
batch production, thereby pre-empting potential issues that dataset employed, however, was based solely on data from
may arise during the production process [3, 10, 16]. secondary sources, which is subject to a high level of incon-
In recent times, machine learning methods, such as multi- sistency given that differences in mixing and the qualities of
ple linear regression (MLR), response surface methodology the materials used to produce the concrete could not be taken
(RSM), fuzzy logic algorithm (FLA), genetic algorithm into consideration.
(GA), support vector regression (SVR), radial basis function Previous studies that used machine learning algorithms
(RBF), Krill Herd Algorithm (KHA), adaptive neuro-fuzzy to model the behaviour of concrete characteristics had to
inference system (ANFIS) and artificial neural networks rely on secondary data. However, variation in the source of
(ANN) have been employed to model the performance data presents difficulties, which include inherent noise in the

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

International Journal on Interactive Design and Manufacturing (IJIDeM)

dataset, longer training periods and discrete training func- where xnor m  normalized value of the input variable,
 Xi
tions. For the model to reflect the relationship between the  actual input variable to be normalized Min X i, n(i...n)
process inputs and CPPs, especially for the purpose of batch  minimum
 value
 in the range of the input variable and
production this variation must be minimized. In light of the Max X i, n(i...n)  minimum value in the range of the input
variation, the current study used experimental data from 103 variable [23].
observations to model, predict, and compare the rheological
and compressive strength characteristics of green concrete n  2
i1 Ŷl − Yi
that contains cement, ground granulated blast furnace slag, R2  1 −  
n 2 (3)
fly ash, water, superplasticizer, coarse aggregates and fine i1 Yi − Y l
aggregates using MLR and ANN models. For both models,
equations that relates the process variables to the perfor- where n  number of observations, Y  actual experimen-
mance parameters were established. Their performance was tal values, Ŷ  predicted values and Y  mean of actual
evaluated using indices such as p value, residual-predicted experimental values.
plots, R2 and MSE.
2.2 Artificial neural network (ANN) method

2 Materials and methods ANN approach is a data processing and modelling method-
ology that has been widely used in various scientific domains
A total of 103 experimental observations with varying con- [17, 22, 26, 27]. It has been used successfully in fields such
tents of cement, slag, fly ash, water, superplasticizer, coarse as learning, classification, generalisation, optimization and
and fine aggregates formed the data set for the predicting the specification analysis [22]. The feed-forward backward prop-
engineering properties of green concrete. Table 1 shows the agation neural network system with a sigmoid hidden neuron
range of datasets used for predictive modelling. and a linear output neuron was used in this investigation. As
indicated in Table 1, ANN models were created using seven
normalised input variables and three target attributes. The
2.1 Multiple linear regression (MLR) method data set is chosen at random for training, validation, and test-
ing in the proportion 70%, 15%, and 15%, respectively. To
Numerous technical or scientific challenges arise when ensure that each iteration returned the same set of random
attempting to investigate the interactions between two or selections, a random number generator was initiated prior
more variables. Regression analysis is a statistical technique to the proportioning. The modelling approaches are based
that is extremely effective in such situations [22]. Equation 1 on Levenberg–Marquardt’s back-propagation learning algo-
shows the general equation of the MLR model with n inde- rithm. The number of neurons in the models was adjusted
pendent variables (X1 , X2 …, Xn) and a single dependent from 1 to 5 and the results were evaluated before choosing the
variable (Y). The residual term ∈ in the MLR model can be network with the optimal solution. ANN Models were cre-
considered as an estimate of the error term in the equation. Y ated using the Neural Network Fitting package in MATLAB
is the predicted dependent variable while X 1 and X (n  2, …, n) (2022a). The performance functions for determining the effi-
are the explanatory variables used in the prediction of the ciency of the ANN models were mean-squared error (MSE)
dependent variable, and ∝0 and ∝i, n n(I 1, ..., n) are the coef- and R2 values. MSE can be calculated using the expression
ficients of an affected unit of change in X as presented in in Eq. (4) [22].
Eq. (1). Table 2 shows the input variables and target variables.
To reduce the possibility of the prediction being affected by 1  2
n
the large range of some of the input variables, normalization MSE  Yi − Ŷl (4)
n
of the variables was implemented using Min–Max normal- i1
isation as per Eq. (2) and performance indicators such as
p value, residual-predicted plots and R2 value were used to Model equations for ANN were derived by using the
assess the performance of the model. The general form of the weight, biases and activation functions as illustrated by [28,
regression equation is presented as Eq. (3) below [23–25]. 29]. The general mathematical expression of the equation is
represented in Eq. (5).

Y(1, 2, 3)M L R ∝0 + ∝1 X 1 + ∝2 X 2 + · · · + ∝n X n + ∈ (1) 

n
Y(1, 2, 3)AN N  [W 2]b×a × [Z ]1×b
T
+ Bias2b (5)
  a1
Xi − Min X i, n(i...n) 
xnor m(i)      (2) 2 2 2
[Z ]  − 1, − 1, . . . , −1 (6)
Max X i, n(i...n) − Min X i, n(i...n) 1 + e−2q1 1 + e−2q2 1 + e−2qn

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

Table 1 Range of datasets used

for predictive modelling Materials Unit Minimum Maximum Average Standard deviation

Cement content kg/m3 137 374 229.89 78.87

GGBS content kg/m3 0 193 77.97 60.46
Fly ash content kg/m3 0 260 149.01 85.42
Water kg/m3 160 240 197.17 20.21
Superplasticizer 4.4 19 8.54 2.81
Coarse aggregates kg/m3 708 1050 883.98 88.39
Fine aggregates kg/m3 641 902 739.60 63.34

Table 2 Variables for modelling the engineering properties of green concrete

Input variables Target variables

Cement Slag Fly ash Water SP Coarse Fine Slump Flow Compressive
content content content content aggregates aggregates (cm) (cm) strength
(kg/m3 ) (kg/m3 ) (kg/m3 ) (kg/m3 ) (kg/m3 ) (kg/m3 ) (MPa)

Actual X1 X2 X3 X4 X5 X6 X7 Y1 Y2 Y3
inputs
Normalized x1 x2 x3 x4 x5 x6 x7 – – –
inputs


n Regular Residual of Sheet1 H"Actual Slump"
[q]1×n  [W 1]b×a × [x]1×b + Bias1b (7) 15
Regular Residual of Actual Slump

a1
10

5
where n  number of observations, Y  actual experimental
values, Ŷ  predicted values, a  number of input variables, 0
b  number of neurons in hidden layer, [W 1]  matrix of the
weights in the input layer, Bias 1and 2  biases in the input -5

and output layers, [W 2]  matrix of the weights in the output -10

layer, [Z]  matrix of the transfer function [p]  matrix of
input conversion function and [x]  matrix of normalized -15

input variables -20

10 20 30
Predicted Slump

3 Results and discussion Fig. 1 Variation between the experimental values and predicted values
of slump using MLR

3.1 Results of MLR predictive model

The MLR analysis of the slump, flow and compressive

strength of green concrete is presented as shown in Table common with regression models. When the number of inde-
3. For the slump, the low values of R2 (0.323) and Adjusted pendent variables ≥ 3, there is the tendency for a significant
R2 (0.273) suggest that MLR was not efficient at predict- model term to masks the truly significant ones thus making
ing this behaviour of green concrete. According to Myers affecting the ability of the model to generalize the relation-
et al. [30], the low values of R2 obtained is not suggestive ship between the response and regressor variables. Although
of an entirely poor model but rather an indication that there the residuals are evenly distributed on either side of the zero
is no linear relationship between the performance parameter margin, as shown in the Fig. 1, the large range of residuals (−
(slump) and the regressor variables x1 , x2 , x3 , … and x7 . 15 to + 15) obtained using MLR supports the inefficiency of
Statistically, Yu and Yao [23] are of the opinion that such MLR model in predicting the workability of green concrete.
occurrence could also be as a result of masking effect which is It was also observed that water and fine aggregates content

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

International Journal on Interactive Design and Manufacturing (IJIDeM)

Table 3 Performance of MLR models for engineering properties of green concrete

Engineering property Regression statistics RMSE Normalized variables p value F Significance F

Y1 -slump (cm) Multiple R 7.420 x1 0.709 6.484 2.66885E−07

0.568 x2 0.320
R square
0.323 x3 0.320
Adjusted R square x4 1.7E−5
0.273 x5 0.504
x6 0.204
x7 0.071
Y2 -flow (cm) Multiple R 0.708 12.850 x1 0.759 16.046 9.2E−14
R square x2 0.667
0.502
Adjusted R square x3 0.667
0.465 x4 2.6E−8
x5 0.883
x6 0.976
x7 0.520
Y3 -compressive strength (MPa) Multiple R 2.610 x1 0.135 131.694 5.036E−46
0.9470 x2 0.135
R square
0.898 x3 0.035
Adjusted R square x4 0.035
0.889 x5 2.5E−36
x6 0.258
x7 9.1E−36

were significant model variables due to their p values which Regular Residual of Sheet1 I"Flow"
was lower than 0.05. 25
The R2 (0.502) and adjusted R2 (0.465) obtained for 20
Regular Residual of Flow

flow behaviour were higher than those obtained for slump 15

10
behaviour, indicating that MLR was able to generalise the
5
relationship between flow properties better than slump prop-
0
erties. However, the high residual values (− 25 to + 25) -5
obtained suggest a less-than-perfect model, as seen in Fig. 2. -10
Extremely low p value of the water content (2.6E−8) indi- -15

cated that it had the most significant individual effect on -20

-25
the flowability of the concrete than the other components of
-30
the green concrete. Results obtained in this study are simi-
-35
lar to the findings of Choi et al. [31] where it was reported
20 25 30 35 40 45 50 55 60 65 70 75 80 85
that numerical unimodal models, like Liu model, Krieger— Predicted Flow
Dougherty model and Chong model all gave low levels of
accuracy when compared to the experimental flow values. To Fig. 2 Variation between the experimental values and predicted values
have a more accurate prediction of the flow properties using of flow using MLR
MLR, the particle shape effects, packing density as well as
shear stress between the concrete and conveying medium
must be considered as part of the process parameters [32]. ash, superplasticizer and fine aggregates had significant indi-
MLR was able to predict the compressive strength with a vidual effects on the compressive strength of the concrete
higher level of accuracy with a low standard error (2.608), due to their very low p values. The low range of residuals
high values of R2 (0.896) and adjusted R2 (0.889). In addi- (− 5 to + 5) in Fig. 3 further substantiates the efficiency of
tion, the regression statistics revealed that cement, water, fly the model in predicting compressive strength. According to
Bal Besikci [33] and Rostampour et al. [34], low values of

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

Table 4 Performance of ANN

models for engineering Engineering property Neurons in hidden layer
properties of green concrete
Performance measure 1 2 3 4 5

Slump R2 Training 0.840 0.443 0.375 0.942 0.884

R2 Validation 0.810 0.549 0.658 0.917 0.974
R2 Test 0.864 0.438 0.038 0.807 0.401
R2 Overall 0.83 0.458 0.405 0.918 0.857
MSE 27.53 89.19 44.96 13.75 6.45
Flow R2 Training 0.833 0.828 0.707 0.893 0.862
R2 Validation 0.796 0.898 0.745 0.762 0.957
R2 Test 0.636 0.754 0.443 0.641 0.529
R2 Overall 0.787 0.826 0.694 0.838 0.848
MSE 96.06 70.70 135.22 146.10 34.52
Compressive strength R2 Training 0.942 0.964 0.997 0.990 0.96
R2 Validation 0.980 0.967 0.992 0.974 0.97
R2 Test 0.912 0.960 0.966 0.966 0.964
R2 Overall 0.941 0.962 0.945 0.985 0.963
MSE 2.496 4.94 1.480 2.85 3.97

Regular Residual of Sheet1 J"Compressive Strength" the normalized input variables to predict the properties of
green concrete.
Regular Residual of Compressive Strength

10

Y1M L R  5.237 + 0.967x1 − 1.499E15x2 + 2.017E15x3

5 + 18.742x4 − 2.779x5 + 6.866x6 + 7.599x7 (8)

Y2M L R  34.829 − 1.376x1 − 1.12E15x2 + 1.51E15x3

0 + 43.523x4 + 1.060x5 + 0.270x6 + 4.671x7 (9)

Y3M L R  85.724 − 3.25E15x1 + 1.164x2 − 1.568x3

− 30.089x4 − 1.659x5 − 38.20x6 − 25.867x7 (10)
-5

15 20 25 30 35 40 45 50 55
3.2 Results of ANN predictive model
Predicted Compressive Strength

Fig. 3 Variation between the experimental values and predicted values Table 4 shows the results of the ANN modelling of the
of compressive strength using MLR slump, flow, and compressive strength of green concrete.
The neurons were varied from 1 to 5 in a bid to obtaining
an optimal network architecture (ONA) within this range
residuals indicate a significant model, whereas higher values for each property being modelled. To be regarded effective,
indicate a poor generalisation of the relationship between the a neural network must find a balance between the R2 val-
independent and dependent variables. ues of the training, validation and testing datasets [16, 20].
Other possible reasons for MLR’s inefficiency in predict- The ANN with 7-4-1 configuration was chosen for the slump
ing the slump and flow properties of green concrete are behaviour due to its reduced MSE (13.75) and minimal vari-
interactive effects of individual materials, non-linearity of ance between the R2 values of the training (0.942), validation
the phenomenon and occurrences of simultaneously inexpli- (0.9174) and testing (0.8073) datasets as shown in Fig. 4.
cable variations, thus making it difficult to develop the most Although the MSE validation of the 7-5-1 network was the
effective model [22]. MLR generalization of the behaviour lowest (6.45), the high variation between the R2 values of the
of slump, flow and compressive strength properties of green training, validation and testing datasets indicates an under-
concrete are represented as Eqs. (5–7). These equations use fitted model that is unable to generalise the behaviour of the

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

International Journal on Interactive Design and Manufacturing (IJIDeM)

Fig. 4 Optimal performance

measures for prediction of slump

concrete for other datasets that are different from the train- able to give a better generalisation of the engineering prop-
ing datasets. According to Amar et al. [17], ANN models erties of green concrete due to its ability to identify intricate
with high values of R2 suggest that the variation between the nonlinear relationships between dependent and independent
experimental and predicted values are minimal. variables and the availability of a wider range of options to
Similar to the slump models are the ANN models for pre- improve the flexibility of the model [26, 33, 36]. Using the
dicting the flowability of green concrete. The ANN with an general form of the ANN equation described in Eqs. (5–7)
ONA of 7-5-1 configuration had the lowest MSE of 34.52. of this study as well as the weights and biases obtained from
However, the ONA of 7-4-1 chosen for predicting the flowa- the ONAs after training (Table 5). Equations (12–14) below
bility of concrete was the one with a minimal variation are the ANN matrix model equations for the slump, flow,
between the R2 values of the training (0.828), validation and compressive strength of green concrete. The input layer
(0.898) and testing (0.754) datasets as well as a low MSE conversion functions [q] are calculated using the expression
of 70.70 as seen in Fig. 5. in Eq. (7).
Table 4 shows that the ANN model for compressive
strength demonstrated very strong generalisation of the rela- ⎡ ⎤ ⎡ ⎤
q1 −0.518 −1.426 −0.803 0.112 −0.7706 −0.108 1.537
tionship between the features and the output parameters for ⎢ ⎥ ⎢ ⎥
⎢ q2 ⎥ ⎢ 0.141 −0.834 0.052 1.2938 0.006 2.322 1.208 ⎥
all variations of neuron. The 7-3-1 ONA presented as Fig. 6 ⎢ ⎥⎢ ⎥
⎣ q3 ⎦ ⎣ 1.013 1.632 2.047 −6.475 −1.890 −2.582 −1.0991 ⎦
had the best performance with a MSE of (1.48) and R2 val- q4 −1.103 0.299 1.148 −2.096 −0.863 −0.178 −1.7156 7×4
ues for training, validation, and testing of 0.997, 0.992, and ⎡ ⎤
x1
0.966, respectively. The findings obtained in this study are ⎢ ⎥ ⎡ ⎤
⎢ x2 ⎥
⎢ ⎥ 2.152
similar to the works of Kumar et al. [16], Amar et al. [17] ⎢ x3 ⎥ ⎢ ⎥
⎢ ⎥ ⎢ −0.204 ⎥
Netam et al. [35] where it was reported that ANN is efficient ×⎢ ⎥
⎢ x4 ⎥ +⎢ ⎥ (11)
⎢ ⎥ ⎣ −3.748 ⎦
⎢ x5 ⎥
in predicting the compressive strength of concrete and is very ⎢ ⎥
⎣ x6 ⎦
−3.933 1×4
accurate with high R2 values. Generally, ANN models were x7 1×7

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

Fig. 5 Optimal performance

measures for prediction of flow


  2 3.3 Comparison of MLR and ANN predictive models
Y1AN N  1.138 −0.334 −1.052 0.691 ×
1 + e−2q1
Figures 7, 8 and 9 show a comparison of experimental results
T
2 2 2 with predictions generated using the equations developed
× −1 −1 −1 −1 in this study. The accuracy of the prediction was evaluated
1 + e−2q2 1 + e−2q3 1 + e−2q4
using Eq. (15) below. For MLR, the level of accuracy for the
+ [−0.895] (12) slump, flow and compressive strength predictions are 51%,
72% and 94%, respectively. Although, the level of accuracy
Similarly, Y2ANN and Y3ANN can be obtained as follows. of the model was slightly above average for slump and flow
but when compared to the results obtained from numerical
 
Y2 AN N  −1.776 0.794 rheological models evaluated by Choi et al. [31], MLR still
 performed better.
T
2 2 On the contrary, ANN models were well fitted to the exper-
× −1 −1
1 + e−2q1 1 + e−2q2 imental data in all three scenarios, as can be seen in the higher
  accuracy values of 81%, 83% and 98% obtained for slump,
+ −0.597 (13)
flow and compressive strength in that order. The failure of
  MLR to predict efficiently lower values of slump and flow
Y3AN N  3.324 −1.136 −0.864 properties as shown in Figs. 7, 8 and 9 is regarded as a lim-
 T itation, given that SF, GGBS, FA, and other supplementary
2 2 2 cementitious materials are known to reduce rheological prop-
× − 1 − 1 − 1
1 + e−2q1 1 + e−2q2 1 + e−2q3 erties when used at high concentrations, owing to their finer
 
+ −0.584 (14)

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

International Journal on Interactive Design and Manufacturing (IJIDeM)

Fig. 6 Optimal performance

measures for prediction of
compressive strength

particle content and texture [35–41]. To have a more accu- 4 Conclusion

rate prediction of the rheologic properties using MLR, the
particle shape effects, packing density as well as shear stress In this study, MLR and ANN were used as predictive tools for
between the concrete and conveying medium must be con- modelling the relationship between process inputs and criti-
sidered as part of the process parameters [31]. By doing so cal performance measures for green concrete production. The
the number of the predictors increases and improves the vari- performance of these techniques was evaluated using statis-
ability between the performance response and the prediction tics indices. The following conclusions have been arrived at:
[42–49]. Therefore, this result further substantiates the effi-
ciency of ANN models over MLR. However, this was not the
case for compressive strength as both models were capable i. The multiple linear regression (MLR) models developed
of predicting it with greater accuracy [31]. for predicting slump, flow, and compressive strength
were found to have R2 values of 0.323, 0.506 and 0.898
  with corresponding accuracies of 51%, 72% and 94%,
1  |E x perimental − Pr ediction|
n
Accuracy  1 − suggesting a fair level of precision in the model predic-
n |E x perimental|
i tions. The inadequate generalisation of the initial two
× 100% (15) properties by MLR was ascribed to the non-existence of
the material properties of the conveying medium which
may have impeded the rheological properties of green
concrete.

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

− 0.895

− 0.597

− 0.584
Biases
Hidden layer-output

0.7942
1.138
− 0.334
− 1.052
0.691
− 1.776

0.864
3.324
1.136
Weights

2.152
− 0.204
− 3.748
− 3.933
− 0.056

− 0.988
0.087
2.272
0.289
Biases

1.5372

1.6041
0.1414
− 1.099
1.208

− 1.715
0.265

− 0.566
− 1.083
Fig. 7 Comparison of predictions of the slump properties for green con-
crete
1.0569
− 0.108
2.322
− 2.582
− 0.178
0.320

− 0.093
− 0.583
− 0.355
0.0993
− 0.770
0.006
− 1.890
− 0.863
− 0.069

0.018
− 0.114
− 0.136
2.4664
0.112
1.293
− 6.475
− 2.096
− 0.068

0.196
− 1.534
− 0.608
2.0477
− 0.803
0.052

1.148
0.146
− 2.068
− 0.198
0.230
0.213

Fig. 8 Comparison of predictions of the flow properties for green con-

crete
− 1.426
− 0.834
1.632
0.299

− 1.588
0.105
− 0.257
− 0.351
− 0.15

ii. ANN models with different network architecture were

Input-hidden layer

more accurate at predicting all the evaluated properties

with R2 values of 0.918, 0.826 and 0.945 as well as
accuracy level of 81%, 83% and 98% for slump, flow,
− 0.518
0.141
1.013
− 1.103
0.063
− 0.561
0.131
− 0.027
− 0.465
Weights
Table 5 Weight and biases for ONA

and compressive strength in that order. The use of ANN

model will be more appropriate where there is a limit on
the availability of critical process parameters.
iii. In both cases, the models performed well in predicting
Compressive strength

the compressive strength of green concrete suggesting

that the process parameters adopted in this study are
sufficient to develop performance-based model for com-
pressive strength during batching of green concrete.
7-4-1

7-2-1

7-3-1
Slump
ONA

Flow

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

International Journal on Interactive Design and Manufacturing (IJIDeM)

12. Dixit, S., et al.: Replacing E-waste with coarse aggregate in archi-
tectural engineering and construction industry. Mater. Today Proc.
56, 2353–2358 (2022)
13. Deep, S., Banerjee, S., Dixit, S., Vatin, N.I.: Critical factors
influencing the performance of highway projects: an empirical eval-
uation. Buildings 12(6), 849 (2022)
14. Dixit, S., Stefańska, A.: Digitisation of contemporary fabrication
processes in the AEC sector. Mater. Today Proc. 56, 1882–1885
(2022)
15. Adebanjo, A.U., Dahunsi, B.I.O., Labiran, J.O.: Effects of concrete
grades on strength characteristics of metakaolin modified recycled
aggregate concrete. Niger. J. Technol. Dev. (2021). https://doi.org/
10.4314/njtd.v18i3.3
16. Naik, S.S., Pandey, S., Pawar, S.N., Shinde, B.H., Prakash, C.: Inno-
vative and interactive methodology for development of geopolymer
mortar using fly ash of agricultural waste briquettes. Int. J. Interact.
Des. Manuf. (2023). https://doi.org/10.1007/s12008-023-01280-1
17. Kumar, V., Kutty, S.R.M., Abd Razak, S.N., Shafiq, N., Adebanjo,
A., Baloo, L., Azougagh, A.-A., Iseni, R.: Exploring the untapped
potentials of oily sludge ash blended with fly ash for geopoly-
mer binder via waste valorisation approach. J. Hazard. Mater. Lett.
Fig. 9 Comparison of predictions of the compressive strength for green
(2023). https://doi.org/10.1016/j.hazl.2023.100076
concrete
18. Gupta, S., Kua, H.W., Koh, H.J.: Application of biochar from food
and wood waste as green admixture for cement mortar. Sci. Total
Environ. (2018). https://doi.org/10.1016/j.scitotenv.2017.11.044
19. Harilal, M., Uthaman, S., George, R.P., Anandkumar, B., Thinaha-
ran, C., Philip, J., Kamachi Mudali, U.: Enhanced anti-microbial
References activity in green concrete specimens containing fly ash, nanophase
modifiers, and corrosion inhibitor. Environ. Prog. Sustain. Energy
1. Adebanjo, A.U., Olonade, K.A., Emmanuel, E.: Carbon dioxide (2019). https://doi.org/10.1002/ep.13102
capture and sequestration: an overview of the challenges, potentials 20. Bisht, R.S., Panigrahi, S.K., Singh, S., Kumar, D., Yadav, S.: Design
an opportunities for cement industry in Nigeria. 2, (2021) analysis of a machine for manufacturing of bricks from industrial
2. Dixit, S., Singh, P.: Investigating the disposal of E-Waste as in waste: simulations and experiments. Int. J. Interact. Des. Manuf.
architectural engineering and construction industry. Mater. Today (2021). https://doi.org/10.1007/s12008-021-00784-y
Proc. 56, 1891–1895 (2022) 21. Dixit, S.: Analysing the impact of productivity in indian trans-
3. Suhendro, B.: Toward green concrete for better sustainable environ- port infra projects. IOP Conf. Ser. Mater. Sci. Eng. 1218(1), 12059
ment. Procedia Eng. 95, 305–320 (2014). https://doi.org/10.1016/ (2022)
J.PROENG.2014.12.190 22. Hossain, K.M.A., Lachemi, M.: Strength, durability and micro-
4. Raj, A., Gyaneshwar, A., Chadha, U., Chadha, A., Asija, A., Abrol, structural aspects of high performance volcanic ash concrete.
A., Priya, S., Alam, A., Joshi, A., Bastola, K., Selvaraj, S.K., Cem. Concr. Res. 37, 759–766 (2007). https://doi.org/10.1016/J.
Karthikeyan, B., Nagalakshmi, R., Chandramohan, V., Hadidi, H.: CEMCONRES.2007.02.014
Green manufacturing via machine learning enabled approaches 23. Kumar, C.V., Sargunan, K., Vasa, J.S.S.K., Jesuraj, V.P., Punitha,
(2022) A., Karthikeyan, R.: Applying ANN–PSO algorithm to maximize
5. Hassooni, S., Ethaib, S.: Evaluation the effect of reuse sewage the compressive strength and split tensile strength of blended self
sludge and sewage sludge ash on concrete for cement replacement. curing concrete on the impact of supplementary cementitious mate-
J. Eng. Sci. Technol. 15 (2020) rials. Int. J. Interact. Des. Manuf. (2022). https://doi.org/10.1007/
6. Dixit, S., Stefańska, A., Singh, P.: Manufacturing technology in s12008-022-00907-z
terms of digital fabrication of contemporary biomimetic structures. 24. Amar, M., Benzerzour, M., Zentar, R., Abriak, N.E.: Prediction
Int. J. Constr. Manag. 1–9 (2021) of the compressive strength of waste-based concretes using arti-
7. Murugesan, T., Vidjeapriya, R., Bahurudeen, A.: Sugarcane ficial neural network. Materials (2022). https://doi.org/10.3390/
Bagasse ash-blended concrete for effective resource utilization ma15207045
between sugar and construction industries. Sugar Technol. (2020). 25. Dixit, S., Stefańska, A., Musiuk, A.: Architectural form finding
https://doi.org/10.1007/s12355-020-00794-2 in arboreal supporting structure optimisation. Ain Shams Eng. J.
8. Hossain, S.K.S., Mathur, L., Roy, P.K.: Rice husk/rice husk ash as 12(2), 2321–2329 (2021)
an alternative source of silica in ceramics: a review (2018) 26. Shah, M.N., Dixit, S., Kumar, R., Jain, R., Anand, K.: Causes
9. Hossain, M.U., Liu, J.C., Xuan, D., Ng, S.T., Ye, H., Abdulla, S.J.: of delays in slum reconstruction projects in India. Int. J. Constr.
Designing sustainable concrete mixes with potentially alternative Manag. 21(5), 452–467 (2021)
binder systems: multicriteria decision making process. J. Build. 27. Dixit, S., Stefańska, A., Musiuk, A., Singh, P.: Study of enabling
Eng. (2022). https://doi.org/10.1016/j.jobe.2021.103587 factors affecting the adoption of ICT in the Indian built environment
10. Adebanjo, A., Kareem, M., Olawuyi, O., Ishola, K., Odefemi, A.: sector. Ain Shams Eng. J. 12(2), 2313–2319 (2021)
Effects of waste steel fibres on the mechanical properties of modi- 28. Alaneme, G.U., Mbadike, E.M., Iro, U.I., Udousoro, I.M., Ife-
fied self compacting concrete. J. Eng. Stud. Res. 28, 7–16 (2022). jimalu, W.C.: Adaptive neuro-fuzzy inference system prediction
https://doi.org/10.29081/jesr.v28i2.001 model for the mechanical behaviour of rice husk ash and periwin-
11. Arora, R., Kumar, K., Dixit, S., Mishra, L.: Analyze the outcome kle shell concrete blend for sustainable construction. Asian J. Civ.
of waste material as cement replacement agent in basic concrete. Eng. (2021). https://doi.org/10.1007/s42107-021-00357-0
Mater. Today Proc. 56, 1877–1881 (2022)

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

 International Journal on Interactive Design and Manufacturing (IJIDeM)

29. Garg, C., Namdeo, A., Singhal, A., Singh, P., Shaw, R.N., Ghosh, 41. Mermerdaş, K., Arbili, M.M.: Explicit formulation of drying and
A.: Adaptive fuzzy logic models for the prediction of compressive autogenous shrinkage of concretes with binary and ternary blends
strength of sustainable concrete. In: Lecture Notes in Networks and of silica fume and fly ash. Constr. Build. Mater. (2015). https://doi.
Systems (2022) org/10.1016/j.conbuildmat.2015.07.074
30. Sadowski, Ł, Piechówka-Mielnik, M., Widziszowski, T., Gardynik, 42. Myers, R.H., Montgomery, D.C., Anderson-cook, C.M.: Response
A., Mackiewicz, S.: Hybrid ultrasonic-neural prediction of the Surface Methodology, 3rd edn. (2009)
compressive strength of environmentally friendly concrete screeds 43. Choi, M.S., Kim, Y.J., Kim, J.K.: Prediction of concrete pumping
with high volume of waste quartz mineral dust. J. Clean. Prod. using various rheological models. Int. J. Concr. Struct. Mater. 8,
(2019). https://doi.org/10.1016/j.jclepro.2018.12.059 269–278 (2014). https://doi.org/10.1007/s40069-014-0084-1
31. Faridmehr, I., Bedon, C., Huseien, G.F., Nikoo, M., Baghban, M.H.: 44. Kwon, S.H., Jang, K.P., Kim, J.H., Shah, S.P.: State of the art on
Assessment of mechanical properties and structural morphology of prediction of concrete pumping. Int. J. Concr. Struct. Mater. (2016).
alkali-activated mortars with industrial waste materials. Sustain- https://doi.org/10.1007/s40069-016-0150-y
ability (Switzerland). (2021). https://doi.org/10.3390/su13042062 45. Bal Beşikçi, E., Arslan, O., Turan, O., Ölçer, A.I.: An artificial
32. Öztürk, O.B., Başar, E.: Multiple linear regression analysis and neural network based decision support system for energy efficient
artificial neural networks based decision support system for energy ship operations. Comput. Oper. Res. 66, 393–401 (2016). https://
efficiency in shipping. Ocean Eng. (2022). https://doi.org/10.1016/ doi.org/10.1016/j.cor.2015.04.004
j.oceaneng.2021.110209 46. Rostampour, V., Motlagh, A.M., Hasan Komarizadeh, M., Sadeghi,
33. Yu, C., Yao, W.: Robust linear regression: a review and comparison M., Bernousi, I., Ghanbari, T.: Using Artificial Neural Network
(2017) (ANN) technique for prediction of apple bruise damage (2013)
34. Chen, Y.: Jump or kink: note on super-efficiency in segmented 47. Netam, N., Palanisamy, T.: Prediction of compressive strength and
linear regression break-point estimation workability characteristics of self-compacting concrete containing
35. Akanbi, O.A., Amiri, I.S., Fazeldehkordi, E.: A machine-learning fly ash using artificial neural network. In: Lecture Notes in Civil
approach to phishing detection and defense (2014) Engineering (2023)
36. Leong, L.Y., Hew, T.S., Ooi, K.B., Wei, J.: Predicting mobile wal- 48. Aslam, S., Herodotou, H., Mohsin, S.M., Javaid, N., Ashraf, N.,
let resistance: a two-staged structural equation modeling-artificial Aslam, S.: A survey on deep learning methods for power load and
neural network approach. Int. J. Inf. Manag. (2020). https://doi.org/ renewable energy forecasting in smart microgrids. Renew. Sustain.
10.1016/j.ijinfomgt.2019.102047 Energy Rev. 144, 110992 (2021). https://doi.org/10.1016/J.RSER.
37. Lawal, A.I., Idris, M.A.: An artificial neural network-based math- 2021.110992
ematical model for the prediction of blast-induced ground vibra- 49. Response Surface Methodology in Engineering Science (2021)
tions. Int. J. Environ. Stud. 77, 318–334 (2020). https://doi.org/10.
1080/00207233.2019.1662186
38. Mahmoodi-Babolan, N., Heydari, A., Nematollahzadeh, A.:
Publisher’s Note Springer Nature remains neutral with regard to juris-
Removal of methylene blue via bioinspired catecholamine/starch
dictional claims in published maps and institutional affiliations.
superadsorbent and the efficiency prediction by response surface
methodology and artificial neural network-particle swarm opti-
Springer Nature or its licensor (e.g. a society or other partner) holds
mization. Bioresour. Technol. (2019). https://doi.org/10.1016/j.
exclusive rights to this article under a publishing agreement with the
biortech.2019.122084
author(s) or other rightsholder(s); author self-archiving of the accepted
39. Dixit, S.: Impact of management practices on construction produc-
manuscript version of this article is solely governed by the terms of such
tivity in Indian building construction projects: an empirical study.
publishing agreement and applicable law.
Organ. Technol. Manag. Constr. 13(1), 2383–2390 (2021)
40. Dixit, S.: Study of factors affecting the performance of construction
projects in AEC industry. Organ. Technol. Manag. Constr. 12(1),
2275–2282 (2020)

123

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]