See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/381488849

Machine Learning Approaches for Predicting Concrete Compressive Strength

Article in Journal of Advanced Research in Civil and Environmental Engineering · May 2024

CITATIONS READS

0 18

1 author:

Jyoti Thapa

4 PUBLICATIONS 0 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jyoti Thapa on 18 June 2024.

The user has requested enhancement of the downloaded file.

 Journal of Advanced Research in Civil and Environmental Engineering
Volume 11, Print Issue 1 - 2024, Pg. No. 9-20
Peer Reviewed Journal
Research Article

Machine Learning Approaches for Predicting

Concrete Compressive Strength
Jyoti Thapa
Former Master Research Scholar, School of Engineering, Pokhara University, Pokhara, Nepal.

I N F O A B S T R A C T

E-mail Id: Concrete compressive strength (CS) plays a crucial role in infrastructure
[email protected] development. Accurate and timely prediction of compressive strength
Orcid Id: is crucial for optimising the performance of structural components. In
https://orcid.org/0009-0000-0963-0955 this study, 776 experimental datasets were collected from past research.
How to cite this article: These datasets were analysed with different machine learning (ML)
Thapa J. Machine Learning Approaches for techniques. The study evaluated the applicability of ML approaches
Predicting Concrete Compressive Strength. J in forecasting concrete strength. The forecasted performance of the
Adv Res Civil Envi Engr. 2024; 11(1): 09-20. regression model was compared with different statistical parameters.
Date of Submission: 2024-02-22 In this study, output performance revealed that the random forest (RF)
Date of Acceptance: 2024-03-11 regression model has good CS prediction capabilities by its R-squared
value of 0.91 followed by k-nearest neighbors (KNN), support vector
machine (SVM), and decision tree (DT) with 0.88, 0.84, and 0.78
respectively. Therefore, this research establishes that the ML approach
has a good capacity to forecast the concrete CS based on the real
database. These predictions approach plays perfect integration into
the construction industry to timely prediction of CS of concrete with
high precision and efficiency.
Keywords: Compressive strength, Concrete, Machine Learning,
Regression Model

Introduction engineering, the majority of the linear and non-linear

characteristics of concrete are typically assessed through
Concrete is one of the principal structural components which
the utilisation of its compressive strength. So far, laboratory
has been widely in the construction industry and played
testing has been conducted for the determination of
a crucial role in modern infrastructure development. The
concrete CS. However, this methodology is time-consuming,
main ingredients of concrete are cement, sand, and coarse
very tedious, and costly. In the laboratory, the selection of
aggregate, and these ingredients are bonded together with
ingredients ratio and its properties, admixture composition,
the use of water. The durability of the concrete depends upon
working methodology, environmental conditions, and
the individual properties of these elements which means
testing equipment conditions and efficiency significantly
the physical and mechanical properties of each ingredient
affect the concrete compressive strength.10,11 Moreover, the
significantly affect the strength of concrete. Compressive
possibility of errors in calculation and human error during
strength (CS) of concrete is a fundamental constituent in the
the whole working duration can increase the uncertainties
construction industry, as it affects the structural integrity
in the outputs of CS. Therefore, the actual prediction of
and performance of various infrastructures. Therefore,
CS in laboratories is very challenging. To overcome the
the compressive strength of concrete plays an important
limitations of traditional laboratory techniques, some
role in stable structural design.1–11 In the field of structural
researchers proposed empirical approaches to determine

Journal of Advanced Research in Civil and Environmental Engineering (ISSN: 2393-8307)

Copyright (c) 2024: Author(s). Published by Advanced Research Publications
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 10

the CS of concrete.4,5 Due to the complex non-linear nature Methodology

of concrete, these empirical approaches are unable to
In this research, 776 concrete compressive strength
forecast the actual strength of concrete. Likewise, the
laboratory testing data were collected from different
independent variable used in these empirical equations
previous research papers.9–25 The statistical description of
cannot be accurately applied to determine the CS of
these datasets is presented in Table 1. Additionally, the
concrete, thus, it significantly decreases the accuracy of
cement (C), water (W), sand (S), coarse aggregate (CA), fly ash
other structural components and their strength. Therefore,
(FA), superplasticizer (SP), and curing time (T) were selected
accurate compressive strength prediction by using the
as independent input variables and the CS of concrete was
real-time database is essential.4–6
chosen as the target variable. Afterwards, the selected
Since the last decade, with the advancement of artificial parameters were preprocessed with cleaning, transforming
intelligence (AI), different ML approaches can be applicable the data, correlation, distribution, and normalization to
to overcome these above-mentioned limitations and maintain the same scale for all selected parameters. The
can determine the accurate CS of concrete with a real preprocessed datasets were divided into training datasets
construction database. Many researchers have explored the (80%) and testing datasets (20%). Firstly, the training datasets
different ML techniques for determining concrete CS based were trained with a support vector machine (SVM) regression
on laboratory testing data. These studies concluded that model. Afterwards, training datasets were trained with
these machine learning techniques have strong correlations the decision tree (DT) ML technique. Furthermore, the
between forecasted and actual concrete compressive k-nearest neighbors (KNN) ML technique was used. Finally,
strength with good prediction performance.4,9,12–15 the robust random forest (RF) regression model was applied
In this study, concrete compressive strength laboratory to train the selected databases. More importantly, these ML
datasets were collected from different research. In previous techniques were trained with their optimal hyperparameters.
research limited researchers compared the prediction After that, all train regression model results were tested
results of different ML approaches. Therefore, this with 20 % datasets. The output results of each model were
study presents the comparative evaluation of different evaluated with different statistical parameters. Finally, the
ML regression models to predict concrete compressive best prediction regression model was established with a
strength. In this study, different statistical indices were used comparison of the performance parameters of each ML
to compare the predicted performance and to establish model. The detailed research flowchart is presented in
the best prediction model for forecasting the concrete CS. Figure 1.
Table 1.Statistical Indices of Selected Features

Indicators/ Coarse Fly Curing Compressive

Cement Water Sand Superplasticizer
Ingredients Aggregate Ash Time Strength
Count 776 776 776 776 776 776 776 776
Mean 300.2 182.1 765.7 1026.9 69.6 4.6 31.4 30.9
Std 98.27 17.96 148.09 142.26 63.54 5.33 27.0 13.4
Min 134.7 140 0.0 410.0 0.0 0.0 1.0 6.27
25% 218.9 168.2 734.9 966.0 0.0 0.0 14.0 20.4
50% 290.1 186 780.5 1028.4 97.0 3.7 28.0 29.8
75% 372.0 192.9 830.0 1086.8 124.8 9.4 28.0 39.8
Max 540.0 238.6 1820.0 1385.2 200.1 28.2 100.0 79.9

ISSN: 2393-8307
 Thapa J
11 J. Adv. Res. Civil Envi. Engr. 2024; 11(1)

Figure 1.Schematic Flowchart for Forecasting Concrete Compressive Strength Using ML Techniques
Correlation Analysis negative correlation with water (W), sand (S), and coarse
aggregate (CA). This relationship indicated that the increase
In machine learning approaches, the multicollinearity in cement and curing time increased the concrete CS. In
establishment and visualisation are the part of data contrast, an increase in sand, coarse aggregate, and water
preprocessing phase. It provides a prime idea regarding content decreases the CS of concrete. In addition, a high
the relationship between the selected input parameters positive correlation exists between FA and SP with a value
and the target variables which is illustrated in Figure 2. The of 0.65 and a high negative correlation between FA and C
range of correlation lies between -1 to 1. A high positive with a value of 0.75. This analysis indicates that all these
correlation between selected parameters is indicated parameters have a good correlation and play a crucial role
with a +1 value and vice versa. Figure 2 illustrates that the in forecasting concrete compressive strength.
target variable CS has a positive correlation with curing
time (T), cement (C), and superplasticizer (SP) but it has a

ISSN: 2393-8307
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 12

Figure 2.Pearson’s Correlation Between Selected Dependant and Independent Features

Figure 3.Box-and-Whisker Plot for Data Distribution and Visualisation

Data Distribution CA, T, and CS. These outliers may reduce the regression
model prediction performance. The removal of outliers
In machine learning, data distribution is part of data
increases the prediction capacity of the selected model,
preprocessing, and it refers to the pattern or spread of data
however, these outlier considerations during the regression
points in a dataset. It gives a quick understanding of the
are more meaningful from civil engineering construction
selected dataset ranges. Data distribution understanding
point of view. Therefore, all outliers of features are selected
plays a crucial role in feature selection, outlier identification,
for regression analysis.
and model selection. Therefore, it significantly impacts
the model generalisation and training, and ultimately, Data Normalization
impacts the accuracy of ML approaches. In this study, In regression analysis, data normalization is a very important
the selected dependent and independent features are step of data preprocessing. This process is essential due
presented in a box-and-whisker plot as shown in Figure to the selected datasets with different dimensions. These
3. This presents the key statistical range of each feature different dimensions should be kept in the same and
with a description of the central tendency of datasets. The uniform scale to increase the forecast performance of
selected features have some outliers in the case of W, S, the regression model. The process involved in making the

ISSN: 2393-8307
 Thapa J
13 J. Adv. Res. Civil Envi. Engr. 2024; 11(1)

same scale for all selected features is termed as scaling or datasets, 5-fold cross-validation shows high accuracy as
rescaling of datasets. In this study, the selected features are compared to others. Therefore, in this study, a 5-folded
in different scales. In Table 1, the value of CS ranges from cross-validation method was used to predict the optimal
6.27 MPa to 79.9 MPa but S ranges from 0 to 1820 kg/m3, hyperparameter and train the selected regression models.
this indicates that the selected features are in different This 5-fold cross-validation was applied in each regression
scales. Therefore, a uniform scale is essential to present model during the training and testing phase. During this
these datasets. In this study, min-max normalization was process, firstly the selected database was randomly split
used to maintain uniformity in the range of all selected into five divisions. After that, the selected ML regression
features. The selected datasets were rescaled by using model was trained with four-folds, and the remaining
the following formula: fold was used for testing the datasets. The output of the
Xn = (X – Xmin)/(Xmax – Xmin) (1) validation result is provided with an average value of five-
fold values. In this process, the outputs of the trained and
Where the actual value of the selected features is denoted tested model show high accuracy and better prediction
by X, rescaled values are represented by Xn. Additionally, performance. In this study, the 5-fold cross-validation
the minimum value of each feature is denoted by Xmin, method is used to determine the best hyperparameters
whereas the maximum value is denoted by Xmax. in each ML model, which are presented in Table 2.
Statistical Analysis of Selected Model Table 2.Hyperparameters in Selected ML
Regression Model
In the context of regression analysis, statistical indices have
commonly been employed to assess the effectiveness of Model Optimal Hyperparameter
regression models. In addition, these indicators have been
applied to compare the prediction performance between SVM ‘C’: 50.0, ‘gamma’: ‘scale’, ‘kernel’: ‘rbf’
the selected different regression models. Therefore, in ‘criterion’: ‘friedman_mse’, ‘min_samples_
this research, the following worldwide used equations split’: 2, ‘max_depth’: 15, ‘min_samples_
DT
are applied to determine and compare the performance leaf’: 1, ‘splitter’: ‘random’, ‘max_features’:
of selected regression ML techniques. In this equation, None
the actual values of selected features are represented KNN ‘n_neighbors’: 5, ‘p’: 1, ‘weights’: ‘distance’
by symbols yia, and corresponding predicted values are
represented by yip, and the ‘n’ in the equation refers to the ‘n_estimators’: 100, ‘max_depth’: 20,
total number of data used in method research in ML models. RF ‘min_samples_leaf’: 1, ‘max_features’: ‘sqrt’,
‘min_samples_split’: 2

Support Vector Machine (SVM)

The SVM is a robust machine-learning approach used
for problem-solving in regression tasks. In addition, this
technique can be used for classification problems as well. It
has a powerful capability for accurate predictions of various
scenarios and dealing with non-linear relationships between
the input features and targeted dependent variables by
using the concept of the hyperplane. In this study, the best
hyperparameters of support vector regression (SVR) models
are presented in Table 2. After that, the train regression
model was tested with test set data.
The comparison between the predicted and actual concrete
CS is illustrated in Figure 4. This comparison reveals that
this SVR model has good prediction accuracy in predicting
Compressive Strength Prediction Models and CS. Also, the correlation between predicted and actual
Results values is presented in Figure 5. This demonstrates that
After data preprocessing, the selected databases were split the R-squared value of this regression model is 0.84 with
into training and testing data. The training datasets were Pearson’s r value of 0.92. These output results depict that
trained with selected different machine learning algorithms this regression model has good capability to forecast the
with their optimal hyperparameters. Based on the selected concrete CS based on real laboratory datasets.

ISSN: 2393-8307
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 14

Figure 4.Comparison Between Actual and Forecasted Outcomes in the SVR Model

Figure 5.Relationship Between Predicted Outcomes of the SVR Model with Actual Value of Concrete CS

Table 3.Forecasted Statistical Indicators of the SVR Model

R2 VAF (%) MAE MRE MSE RMSE RRMSE MAPE
0.84 83.7 3.59 -0.22 27.3 5.22 0.16 11.38

The statistical outputs of this model are demonstrated in a non-parametric approach. This robust ML technique is
Table 3. This SVR model shows statistical metrics namely applied to solve regression problems. In addition, this DT
R2 and VAF with values of 0.84 and 83.7%. In addition, model can be used for classification tasks. This method is
the MAE, MRE, and MSE metrics of this model show the very simple to interpret, visualise, and recognise the output
values of 3.59, -0.22, and 27.3. Furthermore, the statistical results. In this method, the decision-based features are
metrics namely RMSE, RRMSE, and MAPE in this regression represented by the internal node. Likewise, the branch of
model depict the value of 5.22, 0.16, and 11.38 respectively. the tree represents the outcomes of the decision, and the
These outcome results show that the SVR model has a good predicted output is represented by the leaf node. The optimal
capability to forecast the concrete CS. hyperparameters of DT models were optimised based on
Decision Tree (DT) the selected datasets of this research study. These optimal
hyperparameters of DT models are presented in Table 2.
In supervised machine learning, the Decision tree (DT) is

ISSN: 2393-8307
 Thapa J
15 J. Adv. Res. Civil Envi. Engr. 2024; 11(1)

Figure 6.Comparison Between Actual and Forecasted Outcomes in the DT Model

Figure 7.Relationship Between Predicted Outcomes of the DT Model with Actual Value of Concrete CS

Table 4.Forecasted Statistical Indicators of DT Model

R2 VAF (%) MAE MRE MSE RMSE RRMSE MAPE
0.78 78.3 3.75 -1.78 34.53 5.87 0.18 13.38

The relationship between the actual value and the Table 4 illustrates the different statistical evaluation indices
forecasted outcomes of concrete CS in the DT regression of the DT model. This DT model shows statistical metrics
model is illustrated in Figure 6. This comparison indicates namely R2 and VAF with values of 0.78 and 78.4 %. In
that the DT has satisfactory accuracy in forecasting the addition, the MAE, MRE, and MSE metrics of this model
concrete CS. Also, the correlation between predicted and show the value of 3.75, -1.78, and 34.53 respectively.
actual values is presented in Figure 7. This demonstrates Furthermore, the statistical metrics namely RMSE, RRMSE,
that the R-squared value of this regression model is 0.78 and MAPE in this regression model depict the value of 5.87,
with Pearson’s r value 0.89. These output results depict 0.18, and 13.38 respectively. These outcome results show
that the prediction capabilities of this regression model that the concrete CS forecasting ability of the DT model is
have acceptable limits. within acceptable limits.

ISSN: 2393-8307
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 16

K-Nearest Neighbors (KNN) model is illustrated in Figure 8. In addition, the correlation

between predicted and actual values is presented in Figure
K-nearest neighbors (KNN) is a simple and most used
9. This demonstrates that the R-squared value of this
ML method for solving problems in regression tasks. In
regression model is 0.88 with Pearson’s r value 0.94. These
addition, this model can be used for classification tasks.
output results depict that this KNN regression model has
The fundamental concept underlying nearest neighbor
good capability to forecast the concrete CS based on the
techniques involves identifying a specified number of training
laboratory datasets. Also, Table 5 illustrates the different
samples that are in the closest vicinity to the new point and
statistical evaluation indices of the KNN model. This KNN
subsequently making a label prediction based on these
model shows statistical metrics namely R2 and VAF with
samples. This technique has a powerful capability for accurate
values of 0.88 and 87.9 %. In addition, the MAE, MRE,
predictions of various continuous data labels between
and MSE metrics of this model show values of 2.91, 0.68,
the input features and targeted dependent variables. In
and 20.2 respectively. Furthermore, the statistical metrics
this study, optimal hyperparameters of KNN models are
namely RMSE, RRMSE, and MAPE in this regression model
determined based on the selected datasets. The optimal
depict the value of 4.5, 0.14, and 9.74 respectively. These
hyperparameters of KNN models are presented in Table 2.
outcome results show that the KNN model has a good
The relationship between the actual value and the capability to forecast the concrete CS.
forecasted outcomes of concrete CS in the KNN regression

Figure 8.Comparison Between Actual and Forecasted Outcomes in the KNN Model

Figure 9.Relationship Between Predicted Outcomes of the KNN Model with the Actual Value of Concrete
CS

ISSN: 2393-8307
 Thapa J
17 J. Adv. Res. Civil Envi. Engr. 2024; 11(1)

Table 5.Forecasted statistical indicators of the KNN model

R2 VAF (%) MAE MRE MSE RMSE RRMSE MAPE
0.88 87.9 2.91 0.68 20.2 4.5 0.14 9.74

Figure 10.Comparison Between Actual and Forecasted Outcomes in the RF Model

Table 6.Forecasted Statistical Indicators of RF Model

Figure 11.Relationship Between Predicted Outcomes of the RF Model with Actual Value of Concrete CS

R2 VAF (%) MAE MRE MSE RMSE RRMSE MAPE

0.91 91.22 2.66 -0.83 14.7 3.83 0.12 9.14

Random Forest (RF) ensemble approach. This technique has a good prediction
capability with high accuracy and over-fitting control
Random forest (RF) is a versatile and powerful supervised
capabilities. In addition, this method can be used to handle
ML technique. This robust ML technique is applied to solve
missing values, outliers, and high-dimensional datasets.
regression problems. In addition, this RF model can be used
In this study, optimal hyperparameters of RF models are
for classification tasks. It belongs to the family of decision
determined based on the selected datasets. The optimal
tree-based methods and is known for its high accuracy and
hyperparameters of RF models are presented in Table 2.
robustness. Also, this technique combines all the predictions
from different base machine learning algorithms using an

ISSN: 2393-8307
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 18

The relationship between the actual value and the forecasted selected regression shows acceptable results and can be
outcomes of concrete CS in the RF regression model is applicable in real construction projects. However, the best
illustrated in Figure 10. Furthermore, the correlation prediction model selection is very important for researchers
between predicted and actual values is presented in and real construction persons to use in the future for the
Figure 11. This demonstrates that the R-squared value of prediction of the CS of concrete and to incorporate it for
this regression model is 0.91 with Pearson’s r value 0.97. structural integrity. The output results of SVM, DT, KNN,
These outcome results show that the RF model has a good and RF regression models are compared using different
capability to forecast the concrete CS. Also, these figures statistical indicators. The best final model for forecasting
show that the forecasting outcomes of this model have a the concrete CS is selected based on the performance of
good fitting with the actual concrete strength values. these regression models. The comparison between these
Table 6 illustrates the different statistical evaluation indices selected four regression models is presented in Table 7.
of the RF model. This RF model shows statistical metrics In this comparison, firstly four weightings such as 1, 2, 3,
namely R2 and VAF with values of 0.91 and 91.22 %. In and 4 are defined as per the lower to higher performance
addition, the MAE, MRE, and MSE metrics of this model show of each model. Compares the outputs of each model and
values of 2.66, -0.83, and 14.7 respectively. Furthermore, ranks them with respective weightage. After that, add all
the statistical metrics namely RMSE, RRMSE, and MAPE in the weightage for each model and compare them with their
this regression model depict the values of 3.83, 0.12, and total weightage. Table 7 presents the overall performance of
9.14 respectively. These results show that the RF model has all models with their performance weightage. It depicts that
good capability for forecasting the CS of concrete. the RF regression model has the highest weightage followed
by KNN, SVM, and DT. In this study, the RF regression model
Comparison of Results shows good prediction performance as compared to other
In previous sections, the concrete CS prediction outputs regression models. In addition, based on the R-squared and
of different ML models were analysed and compared VAF, the RF regression model displays better prediction
with the actual compressive strength results which were capacity to forecast concrete CS as compared to other
tested in the laboratory. The prediction performance of regression models.
Table 7.Comparison Between Statistical Indices of Selected Regression Models

Regression Model/ Indices SVM DT KNN RF

R2
Value 0.84 0.78 0.88 0.91
Weightage 2 1 3 4
VAF (%) Value 83.7 78.40 87.9 91.22
Weightage 2 1 3 4
MAE Value 3.59 3.75 2.91 2.66
Weightage 2 1 3 4
MRE Value -0.22 -1.78 0.68 -0.83
Weightage 3 1 4 2
MSE Value 27.3 34.53 20.2 14.7
Weightage 2 1 3 4
RMSE Value 5.22 5.78 4.5 3.83
Weightage 2 1 3 4
RRMSE Value 0.16 0.18 0.14 0.12
Weightage 2 1 3 4
MAPE Value 11.38 13.38 9.74 9.14
Weightage 2 1 3 4
Total Weightage 17 8 25 30
Remarks
Rank 3 4 2 1

ISSN: 2393-8307
 Thapa J
19 J. Adv. Res. Civil Envi. Engr. 2024; 11(1)

Conclusion References
In this study, 776 laboratory-tested datasets of concrete 1. Banjara R, Thapa D, Katuwal TB, Adhiakari S. Seismic
compressive strength were collected from different research behaviour of buildings as per NBC 105:1994, NBC
papers. These collected data were comprehensively 105:2020 and IS 1893:2016. Proceedings of the IOE
analysed using four machine learning approaches such as Graduate Conference. 2021;10:1464-71.
SVM, DT, KNN, and RF regression models. A compressive 2. Bhusal B, Paudel S, Katuwal TB. Investigation of
assessment of concrete CS was conducted based on these confinement effects for determining moment curvature
selected previous datasets. It has been observed that and interaction diagram of reinforced concrete column.
the performance of all regression models shows good Tech J. 2020;2(1):81-8.
prediction capabilities. Thus, this study reveals that the ML 3. Chaulagain H, Rodrigues H, Spacone E, Varum H.
approach can be applicable to forecast the CS of concrete Assessment of seismic strengthening solutions for
with high accuracy, less computational time, and very existing low-rise RC buildings in Nepal. Earthq Struct.
limited experimental effort. From this study following key 2015;8:511-39.
conclusions emerged. 4. Chithra S, Kumar SS, Chinnaraju K, Ashmita FA.
The concrete CS is positively affected by the volume of A comparative study on the compressive strength
cement, superplasticizer, and curing period. In contrast, prediction models for high performance concrete
the higher amount of water, sand, coarse aggregate, and containing nano silica and copper slag using regression
fly ash used in concrete reduces the CS of concrete. analysis and artificial neural networks. Constr Build
Mater. 2016;114:528-35.
The outputs of this research study demonstrate that the 5. Chopra P, Sharma RK, Kumar M. Regression models
R-squared value of the RF regression model has the highest for the prediction of compressive strength of concrete
value of 0.91 and lower error followed by KKN, SVM, and with & without fly ash. Int J Latest Trend Eng Technol.
DT with values of 0.88, 0.84, and 0.79 respectively. Based 2014;3:400-6.
on the comparison between all the statistical indicators, 6. Gautam D, Bhattarai A, Rupakhety R. Machine learning
the RF regression model has the highest weightage and and soft voting ensemble classification for earthquake
is followed by the KNN, SVM, and DT regression models. induced damage to bridges. Eng Struct. 2024;303:1-16.
Finally, this research reveals that all selected regression 7. Ghimire N, Chaulagain H. Seismic vulnerability
models have good prediction performance for concrete assessment of reinforced concrete school building in
compressive strength prediction. Therefore, these Nepal. Asian J Civil Eng. 2021;22:249-62.
approaches are beneficial to the timely prediction of 8. Katuwal TB. Comparative evaluation of concrete flexural
concrete compressive strength in real construction projects. strength of river bed and crusher run coarse aggregate
Based on the comparison between these selected regression in Pokhara valley. J Innov Eng Educ. 2019;2(1):221-4.
models, the RF regression model has good prediction 9. Parashar A, Aggarwal P, Saini B, Aggarwal Y, Bishnoi S.
performance with higher accuracy and minimum error Study on performance enhancement of self-compacting
between the actual and forecast outcomes. Therefore, the concrete incorporating waste foundry sand. Constr
RF regression model can be applicable in future research and Build Mater. 2020;251:1-11.
predictions of the concrete CS during the real construction 10. Prabhu GG, Hyun JH, Kim YY. Effects of foundry sand as
period. Moreover, it reduces the testing time, and cost of a fine aggregate in concrete production. Constr Build
the project accordingly. Mater. 2014;70:514-21.
11. Singh G, Siddique R. Abrasion resistance and strength
Acknowledgement properties of concrete containing waste foundry sand
The author is grateful to the researchers whose data sets (WFS). Constr Build Mater. 2012;28:421-6.
have been invaluable to the completion of this research 12. Basar HM, Deveci Aksoy ND. The effect of waste
work. Their accurate collection and documentation have foundry sand (WFS) as partial replacement of sand
significantly contributed to the depth of this research work. on the mechanical, leaching and micro-structural
Source of Funding: None characteristics of ready-mixed concrete. Constr Build
Mater. 2012;35:508-15.
Conflict of Interest: None 13. Mahajan L, Bhagat S. Machine learning approaches
There are no known conflicts of interest with the publication for predicting compressive strength of concrete with
of this research article. fly ash admixture. Res Eng Struct Mater. 2022;9:1-28.

ISSN: 2393-8307
Thapa J
J. Adv. Res. Civil Envi. Engr. 2024; 11(1) 20

14. Song H, Ahmad A, Farooq F, Ostrowski KA, Maślak M,

Czarnecki S. Predicting the compressive strength of
concrete with fly ash admixture using machine learning
algorithms. Constr Build Mater. 2021;308:1-15.
15. Song Y, Zhao J, Ostrowski KA, Javed MF, Ahmad A,
Khan MI, Aslam F, Kinasz R. Prediction of compressive
strength of fly-ash-based concrete using ensemble
and non-ensemble supervised machine-learning
approaches. Appl Sci. 2022;12:1-33.
16. Aggarwal Y, Siddique R. Microstructure and properties
of concrete using bottom ash and waste foundry sand
as partial replacement of fine aggregates. Constr Build
Mater. 2014;54:210-23.
17. Agudelo A, Palacio CA, Monteiro SN, Colorado HA.
Foundry sand waste and residual aggregate evaluated
as pozzolans for concrete. Sustainability. 2022;14(15):1-
25.
18. Bilal H, Yaqub M, Rehman SK, Abid M, Alyousef R,
Alabduljabbar H, Aslam F. Performance of foundry sand
concrete under ambient and elevated temperatures.
Materials. 2019;12:1-24.
19. Guney Y, Sari YD, Yalcin M, Tuncan A, Donmez S. Re-
usage of waste foundry sand in high-strength concrete.
Waste Manag. 2010;30:1705-13.
20. Katuwal TB. Comparative analysis of concrete
compressive strength of river bed and crusher run
coarse aggregate in Pokhara Valley. OODBODHAN.
2018;5:23-6.
21. Katuwal TB. Correlation between concrete compressive
strength and rebound number of river bed and
crusher run coarse aggregate in Pokhara Valley. Tech
J. 2019;1:29-33.
22. Manoharan T, Laksmanan D, Mylsamy K, Sivakumar
P, Sircar A. Engineering properties of concrete with
partial utilization of used foundry sand. Waste Manag.
2018;71:454-60.
23. Mavroulidou M, Lawrence D. Can waste foundry sand
fully replace structural concrete sand? J Mater Cycle
Waste Manag. 2019;21:594-605.
24. Siddique R, Singh G, Singh M. Recycle option for
metallurgical by-product (Spent Foundry Sand) in green
concrete for sustainable construction. J Clean Prod.
2018;172:1111-20.
25. Thiruvenkitam M, Pandian S, Santra M, Subramanian
D. Use of waste foundry sand as a partial replacement
to produce green concrete: mechanical properties,
durability attributes and its economical assessment.
Environ Tech Innov. 2020;19:1-10.

ISSN: 2393-8307

View publication stats