https://doi.org/10.31449/inf.v47i5.3951 Informatica 47 (2023) 41–48   41 
A Neuro-Fuzzy Predictor Trained by an Elitism Artificial Electric 
Field Algorithm for Estimation of Compressive Strength of Concrete 
Structures 
 
  
Ch. Sanjeev Kumar Dash
1
, Sarat Chandra Nayak
2
, Ajit Kumar Behera
1
, Satchidananda Dehuri*
3
 
1
Department of Computer Science and Engineering, Silicon Institute of Technology, Silicon Hills, Patia, Bhubaneswar 
– 751024, India 
2
Department of Artificial Intelligence Engineering & Machine Learning, CMR College of Engineering & Technology, 
Hyderabad-501401, India 
3
Department of Information and Communication Technology, Fakir Mohan University, Vyasa Vihar, Balasore-
756019, Odisha, India. 
E-mails: sanjeev_dash@yahoo.com, saratnayk234@gmail.com, ajit_behera@hotmail.com, satchi.lapa@gmail.com 
 
Keywords: metaheuristics, compressive strength of concrete, artificial neural network, artificial electric field 
algorithm, neuro-fuzzy systems, predictive systems, elitism 
Received: Januar 30, 2022 
Meta-heuristic learning-based forecasting is widely acceptable vis-à-vis manual and statistical 
methods in estimating the compressive strength of concrete structures. However, there is always a 
scope for exploring an effective, automated, and accurate predictor for this domain. This article 
proposes an elitism artificial electric field algorithm-based neuro-fuzzy network (eAEFA+NFN) for 
the prediction of compressive strength of concrete structures. The elitism method helps AEFA 
preserve the best individuals from iteration to iteration, by directly placing best fit particles into the 
population for the next generation and thus strengthening the optimization capacity of AEFA. A single 
hidden layer neural network (SHNN) is used as the base model and its inputs are fuzzified using the 
Gaussian triangular membership function with a degree of membership to different classes. The 
optimal number of input data, hidden neurons, bias, and weights for the hidden layer are decided by 
eAEFA. The model is evaluated on samples from a publicly available dataset with curing ages at 3, 7, 
14, and 28 days. Considering four sample series, the eAEFA+NFN produced an average MAPE of 
0.092073 and ARV of 0.139731 which are better compared to others. The experimental outcomes and 
analysis are in favor of the eAEFA+NFN-based forecasting.   
Povzetek:  Predlagana je meta-hevristična metoda eAEFA+NFN za napovedovanje moči betonskih 
struktur, ki temelji na elitistični AEFTA in nevro-mehkih mrežah z Gaussovo triangulacijo.
1 Introduction 
     Estimation of compressive strength (CS) [1] of 
concrete structures is a contemporary research area in the 
domain of manufacturing and construction engineering. 
The manual approach to solving this is very costly and 
time-consuming. Despite the availability of a large 
number of analytical and statistical models, the 
prediction preciseness is not quite satisfactory. In the last 
twenty years or so, multiple machine learning (ML) 
approaches to solve this problem have gained 
momentum, due to the availability of simulated datasets. 
The above-mentioned models are proven to be good in 
extrapolating data and predicting input-output 
relationships. Artificial neural networks (ANNs) are the 
most popularly used approach for concrete structure-
property prediction. ANN-based [2] methods for 
compression strength prediction were also proposed 
where ANN trained with gradient descent method is 
found generating superior results compared to multiple 
regression analysis. Gradient-based learning methods are 
common approaches for ANN training. The pitfalls 
associated with ANN include imprecise learning, 
lethargic convergence rate, and inclination to local 
minima. Some other difficulties associated with ANN are 
appropriate learning mechanisms, optimal network 
structure, computation of the model, etc. However, there 
is no proper way of finding an optimal ANN and it is still 
a challenging task for researchers. Alongside, the fuzzy 
logic system (FLS) which is capable of handling 
uncertainties and incompleteness associated with real-life 
datasets. ANNs are suitable for dealing with quantitative 
and numeric data, while FLS are capable of handling 
qualitative and symbolic data. Individually both have 
reached a degree of maturity and excelled in solving real-
world problems. Integration of ANN [3] and FL gives a 
synergetic effect as compared to an individual. The 
advantage of ANN [4] learning and fuzzy if-then rules 
with suitable membership functions are hybridized to 
obtain a high degree of accuracy in generating nonlinear 
input-output relationships. The hybrid systems combine 
the learning and connectedness architecture of neural 

42   Informatica 47 (2023) 41–48 S.K. Dash et al. 
 
networks with the human-like logical reasoning 
capability of fuzzy systems and take advantage of both. 
Newly, AEFA [5] has been anticipated as an 
optimization method inspired by the principle of 
electrostatic force. AEFA is based on the strong 
theoretical concept of charged particles, electric field, 
and force of attraction/repulsion between two charged 
particles in an electric field. The learning capacity, 
convergence rate, and acceleration updates of AEFA 
have been established through solving some benchmark 
optimization problems. AEFA starts with random 
solutions, fitness evaluation, reproduction, and updating 
the velocity and position of the particles in the search 
space. The updated solution is then compared with the 
previous one, and the better-fit one is retained.  
The objective of this article is to design a robust 
data-driven ML-based forecasting technique for 
modeling and forecasting the compressive strength of 
concrete structures. The forecast capitalizes on the 
approximation ability of ANN and the reasoning 
capability of FLS to generate a neuro-fuzzy network 
(NFN). The proposed eAEFA is used to design the 
optimal parameters of NFN as well as the size of the 
input and the hidden layers, thus producing a hybrid 
forecast, i.e., eAEFA+NFN. The proposed forecast is 
then used to reveal the hidden nonlinear pattern 
associated with the samples in the dataset and evaluated 
through different error metrics. The new answer is then 
compared to the old one, and the one that fits better is 
kept. However, in the preceding level, known as elite 
solutions, there may be a few good options. These elite 
answers are passed down to the next generation without 
alteration in the elitism mechanism. The worst answers 
are phased out in favour of elite ones. The worst 
solutions are replaced by the elite solutions discovered in 
the preceding generation of any generation. With many 
swarms and evolutionary algorithms, the elitism process 
of replacing the poorest answers with elite ones is 
implemented.  The elitism approach aids AEFA in 
preserving the finest individuals from iteration to 
iteration by immediately introducing the best-suited 
particles into the population for the following generation, 
hence enhancing AEFA's optimization capability. 
The rest of the article is organized into four sections. 
Section 2 discusses related works, Section 3 briefs about 
the methods and dataset, Section 4 summarizes the 
experimental outcomes and discussion, followed by 
concluding remarks in Section 5. 
 
2  Related work 
The background of the research work is presented in 
this Section. In Subsections 2.1 and 2.2, literatures study 
of FLANN as a classifier and predictor is discussed. 
Feature selection and its importance are the focus of 
Subsection 2.3. Differential evolution, a meta-heuristic 
computing paradigm is discussed in Subsection 2.4. Feng 
et al. [6] have used the adaptive boosting technique for 
predicting the compressive strength of concrete. By 
taking 1030 sets of data, they have compared their model 
with other individual machine learning techniques like 
artificial neural network (ANN) and support vector 
machine (SVM).  The proposed approach is superior to 
other models. Deng et al. [7] have used a deep learning 
technique, namely convolution neural networks, for the 
prediction of compressive strength of recycled aggregate 
concrete. After that, the model is developed by softmax 
regression. Finally, their model is considered a new 
method for calculating the strength of recycled concrete. 
Salami et al. [8] have proposed a model by using the 
least square support vector machine (LSSVM) to predict 
the compressive strength of ternary-blend concrete. They 
have also applied Coupled simulated annealing (CSA) to 
LSSVM model for better performance. Kumar et al. [9] 
have applied different machine learning algorithms, such 
as Ensemble Learning (EL), Gaussian Progress 
Regression (GPR), Support Vector Machine Regression 
(SVMR), optimized GPR, SVMR, and EL, to predict the 
compressive strength of Lightweight Concrete (LWC). 
Khursheed et al. [10] have used different machine 
learning techniques such as extreme learning machine 
(ELM), emotional neural network, genetic programming, 
relevance vector machine, and min-max probability 
machine regression to forecast the 28-day compressive 
strength of fly ash concrete. Latif [11] has developed an 
LSTM model to predict concrete compressive strength. 
In his model, three statistical indices were used, namely 
the coefficient of determination (R2), mean absolute 
error (MAE), and root mean square error (RMSE). 
Asteris et al. [12] have proposed a   hybrid ensemble 
surrogate machine learning technique to predict the 
compressive strength of concrete. Their HENSM model 
gives a very high predictive accuracy compared with 
other models. 
 
Güçlüer et al. [13] have used Linear Regression (LR) 
algorithms, Decision Tree (DT), Artificial Neural 
Network (ANN), and Support Vector Machine (SVM) to 
measure concrete compressive strength. They found the 
DT algorithm had the least amount of error and is most 
suitable for use in concrete compressive strength 
estimation. 
Abuodeh et al. [14] have attempted to address the 
ambiguity by applying two deep learning techniques to 
identify the critical material constituents that affect 
ANN. 
Sevim et al. [15] have proposed a prediction model 
to predict the compressive strength of mortar samples. 
For model construction, they have used adaptive-
network-based fuzzy inference systems and artificial 
neural networks (ANN). Finally, they compared the 
results with Multi-Linear Regression. Table 1 shows a 
few popular research results. 
 
Table 1:  Brief discussion of use of AEFA and ANN and 
prediction of concrete structure. 
Ref. 
No. 
Author Model  Purpose  Outcome 
[5] Yadav 
et al. 
AEFA Proving of 
AEFA 
AEFA is the 
best 

Estimation of Compressive Strength of Concrete Structures… Informatica 47 (2022) 41–48   43 
 
algorithm optimization 
algorithm 
[6] Feng et 
al.   
Adaptive 
boosting  
Compressive 
strength test of 
the concrete 
structure  
ANN and SVM 
give better 
result 
[7]  Deng et 
al. 
Deep 
learning 
method   
To find the 
strength of 
recycled 
concrete. 
Gives higher 
precision, high 
efficiency, and 
high 
generalization.  
[8]  Salami 
et al.  
LSSVM, 
LSSVM-
CSA 
To find the 
strength of 
ternary-blend 
concrete 
LSSVM-CSA 
model works 
fine than 
another model 
with 
2
R
 a 
value is 0.954 
[9] Kumar 
et al. 
 GPR, 
SVMR, 
EL, LWC 
To predict the 
compressive 
strength of 
Lightweight 
Concrete  
GPR gives the 
highest 
accuracy 
[10]  Khursh
eed al. 
MPMR, 
RVM, 
GP, ENN 
and ELM 
models. 
To predict the 
compressive 
strength of 
concrete 
MPMR gives 
better results 
than other 
models. 
[11]  Latif  LSTM  To predict 
concrete 
compressive 
strength  
LSTM model 
works fine than 
other models. 
[12] Asteris 
et al. 
Hybrid 
HENSM 
model  
 To predict 
compressive 
strength (CS) 
of concrete   
structure  
HENSM model 
better than 
CML model. 
[13] Güçlüe
r et al. 
 (ANN), 
(DT), 
(SVM) 
and (LR) 
algorithm
s.  
To predict the 
concrete 
compressive 
strength 
Best   
correlation 
coefficient and 
best absolute 
error using DT 
algorithm 
[14] Abuod
eh et al. 
SFS and 
NID 
To find critical 
material 
constituents 
that affect the 
ANN 
ANN with SFS 
and NID gives 
improved 
accuracy than 
other models. 
[15] Sevim 
et al. 
ANN, 
ANFIS 
To estimate 
compressive 
strength using 
the chemical 
composition of 
fly 
GA-based 
ANFIS gives 
better result 
[17] Priyada
rshee et 
al. 
ANN   Compressive 
Strength 
(UCS) of 
Kaolin clay 
ANN model 
gives better 
results than the 
MRA model 
[18] Cao et 
al. 
MOMEM  Estimation of 
the ultimate 
shear strength 
of the soil 
MOMEM is 
significantly 
superior to 
other AI-based 
methods 
[19] Dash et 
al. 
QORA-
ANN 
Prediction of 
cryptocurrency 
QORA-ANN is 
better than 
ANN-GA, 
ANN-DE, 
ANN-PSO 
[20] Sharifi 
et al. 
ANN  Compressive 
strength of the 
mortars 
ANN-based 
model gives 
better results for 
finding 
compressive 
strength of the 
mortars  
[22] Anita 
et al. 
AEFA 
with CSS, 
MOA, 
PSO, and 
GSA 
Stability 
condition 
checking of 
AEFA 
AEFA works 
fine for stability 
condition 
checking 
[23] Nayak 
et al. 
 Extreme 
learning-
AEFA 
For optimizing 
the parameters 
of a neural 
network with a 
single hidden 
layer. 
Generates the 
lowest mean 
absolute 
percentage of 
error (MAPE) 
[24] Anita 
et al. 
AEFA 
compared 
with PSO, 
GA, ABC, 
and GSA. 
AEFA is tested 
for two 
benchmark 
problem that is 
six and fifteen 
generator 
power plant 
systems. 
The 
convergence 
rate is fast in 
case of AEFA  
[25] Behera 
et al. 
AEFA+A
NN 
AEFA + ANN 
model   used to 
predict 
software 
reliability 
datasets 
The proposed 
model  is best 
suitable for 
forecasting 
[26] AL-
Dmour 
et al. 
AEFA Placement of 
phasor 
measurement 
units using an 
optimization 
algorithm 
AEFA is best 
suitable for  
OPP problem 
3 Methods and our model  
This section presents methods like eAEFA and NFN in a 
nutshell and the development of the proposed 
eAEFA+NFN in detail. 
3.1 eAEFA 
AEFA simulates the charged particles as agents and 
measures their strength in terms of their charges [5]. The 
particles are moveable in the search domain through 
electrostatic force of attraction/repulsion among them. 
The charges possessed by the particles are used for 
interaction and the positions of the charges are 
considered the potential solutions to the problem. 
According to AEFA, the particle having the highest 
charge is measured as the best individual, and it attracts 
other particles having inferior charge and moves in the 
search domain. The mathematical justification of AEFA 
is illustrated in [5]. The velocity and position of a particle 
at time instant ‘t’ are updated as per Eqs. (1) and (2), 
respectively. 
( 1) ( ) ( )
d d d
i i i i
Velocity t rand Velocity t acceleration t + =  +
       (1) 
 
( 1) ( ) ( 1)
d d d
i i i
Position t Position t V t + = + +
  (2)                              
Elitism, as previously said, is a technique for passing 
down the greatest persons from generation to generation. 
The system never loses the best individuals discovered 
during the optimization process in this manner. Elitism 
can be achieved by inserting several of the best 
individuals into the next generation's population. Here, 
we simulate a potential solution of SHNN as a charged 
particle and its fitness function as the quantity of charge 
associated with that element. 

44   Informatica 47 (2023) 41–48 S.K. Dash et al. 
 
3.2 NFN 
The proposed method uses an SHNN as the base 
model depicted in Figure 1. An input undergoes a 
fuzzification process using the Gaussian triangular 
membership function as shown in Eq.3. The expanded 
vector intensifies the dimensionality of the input vector 
thus, generates a hyperplane that affords larger 
discrimination ability in the input pattern space. 
2
1
2
( ; , )
ii
i
x small
width
Gaussian x c e 
 −
−

=
,          (3)                                           
                                                
where x = input, c = center of the pattern, and  = 
width of the input pattern. Considering smallest, 
medium, and biggest values as the center of the input 
pattern, a unit in the input pattern has a triangular 
membership function as in Eqs. (4) – (6). 
2
1
2
,1
O ( , , )
ii
i
x small
width
i i i i
x small width e
 −
−

=
,   (4)                                                   
                                           
2
1
2
,2
O ( ,medium , )
ii
i
x medium
width
i i i i
x width e
 −
−

=
, and                                          
                                   (5) 
 
2
1
2
,3
O ( ,big , )
ii
i
x big
width
i i i i
x width e
 −
−

=
.                (6)                                                                                          
The medium value of the 
th
i input vector 
( ) /
ii
big small sizeofinputpattern =−
. Each data is 
used as the input of these membership functions (low, 
medium, high) and the outputs 
( 1... , 1,2,3)
ij
O i N j ==
 are the grades of 
membership. For an input vector 
[ 1, 2,..., ]
T
N
X x x x =
 
and membership of
th
i the input pattern as 
,
()
i j i
mf x
,the 
NM  membership matrix after fuzzification process is 
shown in Eq.7. This matrix is supplied to the SHNN as 
input. 
   (7)
                                
 
 
Figure 1: Single hidden layer neural network. 
To increase its optimization potential, the recently 
created AEFA [21, 24] has been infused with the idea of 
elitism. The elitism approach assists AEFA in preserving 
the greatest individuals from generation to generation by 
immediately inserting the best-fit particles into the 
population for the following generation. The architecture 
of the proposed eAEFA+NFN model is presented in 
Figure 2. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Figure 2: Architecture of eAEFA+NFN. 
 
Output at
th
j hidden neuron is calculated using Eq. 8 
and that of output neuron is calculated using Eq. 9. 
 
1
()
n
j ij i
i
Z f Bias Weight O
=
= + 

     (8)              
1
()
m
oj
i
Z f Bias Weight Z
=
=

                   (9)                    
       
 
This output is compared to the target output and the 
error is calculated as  
i i i
Error Target Estimated =−
. We 
simulate a potential solution of SHNN as a charged 
particle and its fitness function (i.e., error) as the quantity 
of charge associated with that element. The velocity and 
position of a particle are updated as per Eq. 1 and 2. 
3.3 eAEFA+NFN  
This section describes the design of eAEFA+NFN 
[21,22,27] model and then the CS forecasting process. As 
discussed earlier, elitism is a mechanism to retain the 
finest entities from generation to generation. By this 
method, the system never misses the finest individuals 
initiated throughout the optimization procedure. Elitism 
can be done by inserting one or more best individuals 
directly into the population for the subsequent 
generation. The overall eAEFA+NFN process is depicted 
in Figure 3. The process starts with a random initial 
population of solutions. An individual of the population 
represents a potential initial weight and bias of NFN. 
This population and the input samples are fed to the NFN 
model and the fitness is evaluated. Based on the fitness, a 
set of elite solutions are selected. The remainder of the 
population is subject to the regular operators of AEFA. 
At the end of the current generation, the updated and 
original solutions are compared and the better one is 
carried over. Here, the worst solutions are substituted by 
the elite solutions and the process passes into the 
subsequent generation. In this way, the elite solutions are 
carried forward through successive generations. Finally, 
the best solution is preserved and used for testing. The 

Estimation of Compressive Strength of Concrete Structures… Informatica 47 (2022) 41–48   45 
 
high-level eAEFA+NFN process is presented by 
Algorithm 1 and depicted in Figure 3. 
 
 
Figure 3: The eAEFA+NFN-based forecasting process. 
 
Algorithm 1: eAEFA+NFN Training 
1. Initialization of population 
2. Setting Input data 
3. Fuzzification of input data using Eq. 4-6 
4. Normalization of fuzzified data 
5. Supply input data and population to the NFN 
6. Apply eAEFA for search updating 
7. Supply test data and the best particle to the NFN and 
preserve the estimated CS value 
4 Experimental data 
The experimental data collected from an open 
repository contains 1030 samples each of which has 9 
real attributes [28]. The attributes 1 – 8 are used as input 
features and the last one as the response variable. All 
instances are quantitative and have numeric values only. 
From the dataset samples curing ages at 3, 7, 14, and 28 
days only are extracted and used for modeling. A 
statistical summary of the dataset is given in Table 2.  
 
Table 2: Summary statistics from the dataset. 
Compo
nent 
Mean Std Min 25% 50% 75% Max 
Cement 
(kg/m3) 
281.167 104.51 102.0 192.3 272.9 350.0 540 
Blast 
furnace 
slag 
(kg/m3) 
73.8958 86.279 0.000 0.000 22.0 142.95 359.4 
Fly ash 
(kg/m3) 
54.1883 63.997 0.000 0.000 0.0 118.30 200.1 
Water 
(kg/m3) 
181.567 21.354 121.8 164.9 185 192.0 247 
Super 
plastici
zer 
(kg/m3) 
6.2046 5.9738 0.000 0.000 6.40 10.20 32.20 
Coarse 
Agg. 
(kg/m3) 
972.918 77.754 801.0 932.0 968 1029.4 1145 
Fine 
Agg. 
(kg/m3) 
773.580 80.176 594.0 730.9 779.5 824.0 992.6 
Age 
(numeri
c) 
45.6621 63.167 1.000 7.00 28.0 56.0 365 
Compre
ssive 
Strengt
h 
(MPa) 
35.8179 16.706 2.330 23.71 34.44 46.135 82.6 
 
Table 3: Error statistics from four sample series and 
seven forecast. 
Forec
ast 
Sample series 
3-days 7-days 14-days 28-days 
MA
PE 
AR
V 
MA
PE 
AR
V 
MA
PE 
AR
V 
MA
PE 
AR
V 
eAEF
A+N
FN 
0.06
275
5 
0.05
1725 
0.0
652
75 
0.05
268 
0.07
753 
0.27
205 
0.16
273 
0.1
82
47 
AEF
A+N
FN 
0.06
530
2 
0.08
2204 
0.1
528
47 
0.05
8725 
0.07
9201 
0.44
5177 
0.30
066 
0.1
82
73 
GD+
NFN 
0.08
845
3 
0.27
4423 
0.3
673
77 
0.29
872 
0.16
439 
0.46
283 
0.32
005 
0.1
87
25 
ANFI
S 
0.08
715
7 
0.29
7405 
0.0
963
57 
0.29
953 
0.09
556 
0.25
293 
0.33
165 
0.2
01
16 
MLP 0.45
440
0 
0.38
9472 
0.4
065
48 
1.00
523 
0.20
604 
0.55
729 
0.35
207 
0.3
30
23 
SVM 0.48
845
2 
0.59
7655 
0.6
864
50 
1.10
308 
0.76
583 
1.02
734 
0.72
005 
0.6
30
47 
MLR 0.89
299
1 
0.93
2005 
0.9
720
35 
1.34
125 
0.92
884 
1.27
026 
1.35
266 
0.9
38
20 
4.1 Experimental outcomes and model 
evaluation 
As stated earlier, we considered only the samples 
with curing ages of 3, 7, 14, and 28 days. The time-series 
approach is used for modeling the data. Model inputs are 
selected using a rolling window method from a sample 
series and normalized using the sigmoid method [29, 30]. 
Two error metrics, mean absolute percentage of error 
(MAPE) and average relative variance (ARV) are used 
for model evaluation and are shown in Eq. 10 -11. 

46   Informatica 47 (2023) 41–48 S.K. Dash et al. 
 
 
1
ˆ 1
100%
N
ii
i i
xx
MAPE
Nx
=
−
=

                           (10) 
2
1
2
1
ˆ()
ˆ()
N
ii
i
N
i
i
xx
ARV
xx
=
=
−
=
−


                                              (11) 
 
To ensure the performance of the eAEFA+NFN 
forecast, four comparative models as NFN with gradient 
descent-based training (GD+NFN), NFN with AEFA 
(AEFA+NFN) adaptive neuro-fuzzy inference system 
(ANFIS), multilayer perceptron (MLP), support vector 
machine (SVM), and multiple linear regression (MLR) 
are also implemented and evaluated with the same input 
patterns. Error statistics from different data series using 
different models are summarized in Table 3. 
 
The best statistics are highlighted in boldface. From 
Table 3, the proposed eAEFA+NFN is found to be the 
best performing and MLR is the least performing 
forecast among all. For the 3-day sample set, 
eAEFA+NFN produced 0.062755 MAPE and 0.051725 
ARV. In the case of the 7-day sample, it achieved 
0.065275 MAPE and 0.05268 ARV. For 14-day samples, 
the MAPE and ARV by eAEFA+NFN are 0.07753 and 
0.27205 respectively. Similarly, in the case of a 28-day 
sample, the proposed model generated the lowest MAPE 
of 0.16273 and ARV of 0.18247. GD+NFN and ANFIS 
are found to be similar in performance. Similarly, the 
performance of MLP and SVM are found closer to each 
other. Overall, the eAEFA+NFN model produced the 
lowest error metric values, offering consistent and 
satisfactory results compared to others. The forecast plots 
shown in Figures 4–7 reflect the accuracy of 
eAEFA+NFN. The model estimations are closer to the 
actuals.  
 
Figure 4: eAEFA+NFN forecast vs actual 
compressive strength values from the 3-day sample series 
 
 
Figure 5: eAEFA+NFN forecast vs actual compressive  
strength values from the 7-day sample series. 
 
Figure 6: eAEFA+NFN forecast vs actual compressive  
strength values from the 14-day sample series. 
 
Figure 7: eAEFA+NFN forecast vs actual compressive  
strength values from the 28-day sample series. 
 
 
Figure 8: MAPE comparison of eAEFA+NFN and 
AEFA+NFN. 
 
Further, to realize the benefit of elitism, we 
compared the MAPE from eAEFA+NFN and that of 
AEFA+NFN. The outcome of the comparative studies is 
depicted in Figure 8. It is observed that in the case of 3-
days and 14-day time series data the performances of 
both models differ slightly from each other while there is 
a significant difference in the case of 7-days and 28-day 
datasets. Similar observations are inferred while 
comparing ARV from both methods. These shreds of 
evidence are in support of the elitism concept and thus 
eAEFA is found better in training NFN. 

Estimation of Compressive Strength of Concrete Structures… Informatica 47 (2022) 41–48   47 
 
5 Conclusions 
A robust hybrid forecast called eAEFA+NFN is 
proposed in this article for accurate and effective 
modeling of compressive strength of concrete cement 
data and forecasting the CS values of unseen samples. 
NFN inputs are fuzzified using the Gaussian triangular 
membership function with a degree of membership to 
different classes. The fuzzified input vectors intensified 
the dimensionality of the input thus, generated a 
hyperplane that affords a larger discrimination ability in 
the input pattern space. The model parameters are fine-
tuned by eAEFA. The enhanced search ability of eAEFA 
and improved approximation ability of NFN combined to 
make the model robust helped in capturing the 
nonlinearity associated with the data. Considering four 
sample series, the eAEFA+NFN produced an average 
MAPE of 0.092073 and ARV of 0.139731 which are 
better compared to others. From comparative studies, it is 
found that the proposed eAEFA+NFN outperformed 
others. Further, the efficiency of the proposed approach 
may be evaluated in other areas of the predictive system. 
The current work may be extended using similar datasets 
from the manufacturing engineering domain. 
 
Acknowledgment: This work is partly supported 
by TARE Fellowship awarded to Prof. Satchidananda 
Dehuri, Department of Computer Science, F. M. 
University, Balasore by SERB, Govt. India vide Sanction 
Order No. TAR/2021/000065.    
References 
[1] Asteris, P. G., & Mokos, V. G. (2019).  Concrete 
compressive strength using artificial neural 
networks. Neural Computing and Applications, 
pp.1-20. 
[2]  Pandey, S., Kumar, V., & Kumar, P. (2021). 
Application and Analysis of Machine Learning 
Algorithms for Design of Concrete Mix with 
Plasticizer and without Plasticizer. Journal of Soft 
Computing in Civil Engineering, 5(1), 19-37. 
[3]  Nayak, S. C., Nayak, S. K., & Panda, S. K. (2021). 
Assessing Compressive Strength of Concrete with 
Extreme Learning Machine. Journal of Soft 
Computing in Civil Engineering, 5(2), 68-85. 
[4]  Nayak, S. C., & Misra, B. B. (2019). A chemical-
reaction-optimization-based neuro-fuzzy hybrid 
network for stock closing price prediction. Financial 
Innovation, 5(1), 1-34. 
[5]  Yadav, A. (2019). AEFA: Artificial electric field 
algorithm for global optimization. Swarm and 
Evolutionary Computation, 48, 93-108. 
[6]  Feng, D. C., Liu, Z. T., Wang, X. D., Chen, Y., 
Chang, J. Q., Wei, D. F., & Jiang, Z. M. (2020). 
Machine learning-based compressive strength 
prediction for concrete: An adaptive boosting 
approach. Construction and Building 
Materials, 230, 117000. 
[7]  Deng, F., He, Y., Zhou, S., Yu, Y., Cheng, H., & 
Wu, X. (2018). Compressive strength prediction of 
recycled concrete based on deep 
learning. Construction and Building Materials, 175, 
562-569. 
[8]  Salami, B. A., Olayiwola, T., Oyehan, T. A., & 
Raji, I. A. (2021). Data-driven model for ternary-
blend concrete compressive strength prediction 
using machine learning approach. Construction and 
Building Materials, 301, 124152. 
[9]  Kumar, A., Arora, H. C., Kapoor, N. R., 
Mohammed, M. A., Kumar, K., Majumdar, A., & 
Thinnukool, O. (2022). Compressive strength 
prediction of lightweight concrete: machine 
learning models. Sustainability, 14(4), 2404. 
[10] Khursheed, S., Jagan, J., Samui, P., & Kumar, S. 
(2021). Compressive strength prediction of fly ash 
concrete by using machine learning 
techniques. Innovative Infrastructure 
Solutions, 6(3), 1-21. 
[11] Latif, S. D. (2021). Concrete compressive strength 
prediction modeling utilizing deep learning long 
short-term memory algorithm for a sustainable 
environment. Environmental Science and Pollution 
Research, 28(23), 30294-30302. 
[12] Asteris, P. G., Skentou, A. D., Bardhan, A., Samui, 
P., & Pilakoutas, K. (2021). Predicting concrete 
compressive strength using hybrid ensembling of 
surrogate machine learning models. Cement and 
Concrete Research, 145, 106449. 
[13]  Güçlüer, K., Özbeyaz, A., Göymen, S., & 
Günaydın, O. (2021). A comparative investigation 
using machine learning methods for concrete 
compressive strength estimation. Materials Today 
Communications, 27, 102278. 
[14] Abuodeh, O. R., Abdalla, J. A., & Hawileh, R. A. 
(2020). Assessment of compressive strength of 
Ultra-high-performance concrete using deep 
machine learning techniques. Applied Soft 
Computing, 95, 106552. 
[15] Sevim, U. K., Bilgic, H. H., Cansiz, O. F., Ozturk, 
M., & Atis, C. D. (2021). Compressive strength 
prediction models for cementitious composites with 
fly ash using machine learning 
techniques. Construction and Building 
Materials, 271, 121584. 
[16] Haykin, S. (2010). Neural networks and learning 
machines, 3/E. Pearson Education India. 
[17] Priyadarshee, A., Chandra, S., Gupta, D., & Kumar, 
V. (2020). Neural Models for Unconfined 
Compressive Strength of Kaolin clay mixed with 
pond ash, rice husk ash and cement. Journal of Soft 
Computing in Civil Engineering, 4(2), 85-102. 
[18]  Cao, M. T., Hoang, N. D., Nhu, V. H., & Bui, D. T. 
(2020). An advanced meta-learner based on 
artificial electric field algorithm optimized stacking 
ensemble techniques for enhancing prediction 
accuracy of soil shear strength. Engineering with 
Computers, 1-23. 
[19] Dash, C. S. K., Behera, A. K., Nayak, S. C., & 
Dehuri, S. (2021, April). QORA-ANN: Quasi 
Opposition Based Rao Algorithm and Artificial 
Neural Network for Cryptocurrency Prediction. In 

48   Informatica 47 (2023) 41–48 S.K. Dash et al. 
 
2021 6th International Conference for Convergence 
in Technology (I2CT) (pp. 1-5). IEEE. 
[20] Sharifi, Y., & Hosainpoor, M. (2020). A predictive 
model-based ANN for compressive strength 
assessment of the mortars containing metakaolin. 
Journal of Soft Computing in Civil Engineering, 
4(2), 1-12. 
 [21] Yadav, A., & Kumar, N. (2019, December). 
Application of artificial electric field algorithm for 
economic load dispatch problem. In International 
Conference on Soft Computing and Pattern 
Recognition (pp. 71-79). Springer, Cham. 
  
 [22] Yadav, A. (2019). AEFA: Artificial electric field 
algorithm for global optimization. Swarm and 
Evolutionary Computation, 48, 93-108. 
[23] Nayak, S. C., Nayak, S. K., & Dash, A. (2021, 
November). Extreme learning with Artifical 
Electric Field Algorithm (EL-AEFA) for estimation 
of Short-term Vehicular Traffic System. In 2021 
IEEE 8th Uttar Pradesh Section International 
Conference on Electrical, Electronics and Computer 
Engineering (UPCON) (pp. 1-4). IEEE. 
[24] Yadav, A., & Kumar, N. (2020). Artificial electric 
field algorithm for engineering optimization 
problems. Expert Systems with Applications, 149, 
113308. 
[25] Behera, A. K., Panda, M., Nayak, S. C., Dash, C., & 
Kumar, S. (2022). An Artificial Electric Field 
Algorithm and Artificial Neural Network-Based 
Hybrid Model for Software Reliability Prediction. 
In Computational Intelligence in Data Mining (pp. 
271-279). Springer, Singapore. 
 [26] Al-Khraisat, B., AL-Dmour, A. S., & Al-Maitah, K. 
(2021, November). Artificial Electric Field 
Algorithm for Optimum PMU Placement. In 2021 
IEEE Green Energy and Smart Systems Conference 
(IGESSC) (pp. 1-6). IEEE. 
[27] Nayak, S. C., Dash, C. S. K., Behera, A. K., & 
Mishra, B. B. (2021, December). A machine 
learning approach for estimating compressive 
strength of concrete structures using an artificial 
electric field algorithm-based neuro-fuzzy predictor. 
In 2021 19th OITS International Conference on 
Information Technology (OCIT) (pp. 229-233). 
IEEE. 
[28] http://archive.ics.uci.edu/ml/datasets/Concrete+ 
Compressive+Strength. 
[29] Nayak, S. C., Misra, B. B., & Behera, H. S. (2019). 
ACFLN: artificial chemical functional link network 
for prediction of stock market index. Evolving 
Systems, 10(4), 567-592. 
[30] Nayak, S. C., Misra, B. B., & Behera, H. S. (2014). 
Impact of data normalization on stock index 
forecasting. International Journal of Computer 
Information Systems and Industrial Management 
Applications, 6(2014), 257-269.