Research on prediction of compressive strength of fly ash and slag mixed concrete based on machine learning

Meng Wang; Jiaxu Kang; Weiwei Liu; Jinshuai Su; Meng Li

doi:10.1371/journal.pone.0279293

Abstract

Every year, a large amount of solid waste such as fly ash and slag is generated worldwide. If these solid wastes are used in concrete mixes to make concrete, it can effectively save resources and protect the environment. The compressive strength of concrete is an essential indicator for testing its quality, and its prediction is affected by many factors. It is difficult to predict its strength accurately. Therefore, based on the current popular machine learning supervised learning algorithms: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVR), three models established a nonlinear mapping between multi-factor features and target feature concrete compressive strength. Using the three completed training models, we validated the test set with 206 example sets, and the Root Mean Square Error (RMSE), fitting coefficient (R²), and Mean Absolute Error (MAE) were used as evaluation metrics. The validation results showed that the values of RMSE, R², and MAE for the RF model were 0.1, 0.9, and 0.21, respectively; the values of XGBoost model were 0.05, 0.95, and 0.15, respectively. The values of SVR were 0.15, 0.86, and 0.3, respectively. As a result, Extreme Gradient Boosting (XGBoost) has better generalization ability and prediction accuracy than the other two algorithms.

Citation: Wang M, Kang J, Liu W, Su J, Li M (2022) Research on prediction of compressive strength of fly ash and slag mixed concrete based on machine learning. PLoS ONE 17(12): e0279293. https://doi.org/10.1371/journal.pone.0279293

Editor: André Gustavo de Sousa Galdino, Instituto Federal do Espírito Santo: Instituto Federal de Educacao Ciencia e Tecnologia do Espirito Santo, BRAZIL

Received: September 14, 2022; Accepted: December 4, 2022; Published: December 27, 2022

Copyright: © 2022 Wang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data presented in this study are openly available in [FigShare] reference number [https://doi.org/10.6084/m9.figshare.21493236.v1].

Funding: This work was supported by the local technical service projects in 2020 from Liaoning Education Department (serial number:20-1114). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

1. Introduction

Concrete is widely used in the construction industry because of its excellent performance [1]. Every year the whole world, because of the construction needs, the amount of use of concrete will be huge [2]. Thus, the carbon footprint associated with cement, and therefore conventional concrete, production is high. Increasing the use of supplementary cementitious materials (SCMs) in concrete is an obvious and necessary step to reduce carbon emissions [3, 4]. Mohamed and Tayeb [5] have studied that concrete is generally mixed with cement, sand, stone, water, and other materials by a certain proportion of the formation of a rigid material. SCMs, such as fly ash and slag are often waste materials from industrial processes. The wide application of these solid wastes in concrete has become an effective measure to reduce the cement consumption and delay the hydration heat, its application not only has a good economy but also can make the concrete have better mixing performance, hardening properties, and durable performance [6–8]. Inclusion of SCMs in concrete lessens the environmental impact of concrete in several ways, in that it: (1) reduces cement consumption and thereby production, (2) can reduce the amount of inert filler (typically sand in conventional concrete) required and (3) uses waste materials that would otherwise be landfilled.

In the current studies, the machine learning-based model has been widely used in slope stability prediction [2], floods [9], prediction of mechanical properties of materials [10–12] and building structures [13, 14]. In addition, many researchers have explored the application of machine learning in concrete prediction of compressive strengths. For example, the studies of Yeh et al. [15–17], built an ANN model with a backpropagation algorithm to predict the compressive strength of concrete at different ages (3 days, 14 days, 28 days, and 90 days). Fan and Chiong et al. [18] predicted the compressive strength of concrete with a support vector machine by studying the composition of concrete. The experimental results show that the model has good performance for the reverse prediction of concrete members under the conditions of multiple input, single output and multiple input and multiple output. Behnood and Golafshani [19] used decision trees to predict the compressive strength of concrete with fly ash and other waste materials; The results indicated that the proposed models could provide reliable predictions of the target mechanical properties. Deshpande and Londhe [20] estimated the compressive strength of recycled aggregate concrete with the artificial meridian; The results indicate that ANN learns from the examples and grasped the fundamental domain rules governing strength of concrete. At present, the research on making concrete with solid wastes such as fly ash and slag as mixing material needs to be further improved [21]. Because of the small sample size of previous studies, the over-fitting problem is easy to occur in model training. The data set samples used in this study reached 1030 groups, which has a better optimization for the over-fitting problem. Using limited data to predict the compressive strength of this type of concrete will help realize engineering applications. At the same time, compared with previous models, the application of a more advanced XGBoost model to a concrete compressive strength test is also helpful to improve the prediction accuracy and speed; In previous experiments also have the problem of lack of field experiment data validation, theoretical research should return to the practical application, most of these studies only compared to the existing data sets, without actually making a concrete test block, a lack of practicality, to solve this problem, after completing the model performance comparison, concrete test blocks will be made on-site. The compressive strength test is carried out after curing for a fixed time in the constant temperature curing box to verify the model’s accuracy.

2. Research methods and steps

The mechanical properties of concrete under the action of multi-feature coupling are affected by many factors, which overlap and influence each other, showing a very complex nonlinear relationship [22]. In this paper, the prediction accuracy and practicability of the three machine learning models, XGBoost model, SVR model, and RF model, will be evaluated by comparing the curve fitting degree between the predicted value and the actual value of the three machine learning models, as well as statistical indicators such as RMSE, R², and MAE, as well as experimental verification. The programming language used was Python3.9, programming analysis on the Notebook platform.

To better conduct model training and model verification, this study will follow the ideas shown in Fig 1, mainly including data processing, model training, model testing, model comparison, and experimental validation. The specific steps are as follows:

Data preprocessing, the data were first preprocessed, such as quality control and dataset partitioning, and correlation analysis.
In machine learning model construction, first build three models of RF, XGBoost, and SVR. After obtaining the initial model, use the training set samples to train the model, then use the divided test set samples to test the three models, and optimize through grid search parameters, so that the model performance is optimal. Finally, the model with the optimal parameters is selected [23].
Comparison of model prediction performance. By inputting the test set into three models and obtaining the model predictive value, draw the fitting curve, scatter chart, relative error chart and Taylor chart between the actual value and the predictive value, and calculate Root Mean Square Error (RMSE), coefficient (R²), Mean Absolute Error (MAE). Then the prediction performance of the model is evaluated according to the graph and statistical indicators, and the model with the best comprehensive performance is obtained [24].
Example verification of optimal model. Make a concrete test block with the same characteristics as the data set, put the completed test block into a constant temperature curing box for 28 days, perform a compressive strength test on the cured concrete test block and record the experimental data; The values of the parameters are input into the optimal model, and the compressive strength is predicted; the average error rate is calculated by comparing the experimental data and the expected data, to test the performance of the model in the actual working state [25].

Download:

Fig 1. Model training and experimental verification process.

https://doi.org/10.1371/journal.pone.0279293.g001

3. Data analysis

3.1 Sample data acquisition and preprocessing

The sample data were based on the Concrete Compressive Strength dataset [26], provided by Professor Yicheng Ye, Department of Information Management, Chunghua University, Taiwan [27] (Due to the large amount of data can not be fully displayed, only part of the data is shown here, the complete data is detailed in this data warehouse: https://doi.org/10.6084/m9.figshare.21493236.v1). The sample dataset consisted of 1030 sets of samples. Each of these had nine features: cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate, age, and concrete compressive strength. The first eight features are input features, and the ninth feature is the target feature. Eight hundred twenty-four groups of samples are randomly assigned as training samples, and the remaining 206 groups are test samples. The specific mix proportion and strength test values of some samples are shown in Table 1; The relevant statistical description of the data set is shown in Table 2.

Download:

Table 1. Experimental data.

https://doi.org/10.1371/journal.pone.0279293.t001

Download:

Table 2. Overall statistical description of data.

https://doi.org/10.1371/journal.pone.0279293.t002

3.2 Correlation analysis of sample data

Before model training, samples’ features unrelated to concrete compressive strength should be excluded, so correlation analysis should be conducted. Correlation analysis needs to verify the distribution state of sample data; the histogram and normal curve of each characteristic value distribution is shown in Fig 2. There are three commonly used correlation coefficients: Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient. The two variables using the Pearson correlation coefficient should follow a normal distribution and have a linear correlation trend; the Pearson correlation coefficient is used to measure the degree of linear correlation [28]. The value range is [–1,1]. The larger the absolute value of the coefficient is, the stronger the correlation is; the closer the absolute value is to 0, the weaker the correlation is. As shown in Fig 2, Fig 2(A), 2(D), 2(F) and 2(G) follow a normal distribution with Fig 2(I), so the Pearson correlation coefficient is used to determine their correlation. The Pearson correlation coefficient is defined as follows: (1)

Download:

Fig 2. Characteristic numerical distribution.

(a) cement; (b) blast furnace slag; (c) fly ash; (d) water; (e) superplasticizer; (f) coarse aggregate; (g) fine aggregate; (h) age; (i) compressive strength of concrete.

https://doi.org/10.1371/journal.pone.0279293.g002

In the Eq (1): E is the mathematical expectation; D is the variance; is the standard deviation; Cov(X,Y) is the covariance of the sum of the random variables, which measures the overall error between the two variables; ρ_XY is the value of the quotient between the covariance and standard deviation between the two variables, also known as the correlation coefficient between the variables X and Y. Variables X and Y in Eq (1) refer to two independent bodies of evidence [29].

If two variables do not obey normal distribution, the spearman correlation coefficient should be used to study their relationship. In Fig 2(B), 2(C), 2(E), 2(H) and 2(I) does not obey the normal distribution, so the Spearman correlation coefficient is used as the judgment basis of correlation. The Spielman correlation coefficient calculation expression is as follows [30, 31]: (2)

In the Eq (2): ρ is the Spearman correlation coefficient, between -1 and 1, |ρ| more close to 1, the greater the relevance; D is the difference between the two data sequences; N is the number of data.

The correlation diagram calculated by Pearson correlation coefficient and Spearman correlation coefficient is shown in Fig 3. In Fig 3, Fig 3(A) is calculated from the Spearman correlation coefficient, and The Pearson correlation coefficient calculates Fig 3(B). It can be seen that the compressive strength of concrete is positively correlated with cement, blast furnace slag, superplasticizer, and age. The compressive strength of concrete is negatively correlated with coarse aggregate, fine aggregate, fly ash, and water. Among them, the compressive strength of concrete is more correlated with cement, water, superplasticizer, and age, and there are no unrelated features. The correlation diagram of each feature is realized by the Heatmap function of the Seaborn library in Python.

Download:

Fig 3. Correlation between features.

(a) Spearman correlation coefficient diagram; (b) Pearson correlation coefficient diagram.

https://doi.org/10.1371/journal.pone.0279293.g003

3.3 Normalization processing

There are nine features in the selected dataset, and the dimensions of the nine features are different. In order to improve the accuracy and stability of the model, reduce the amount of calculation, and achieve the best prediction effect, it is necessary to normalize the data in the dataset. Because the dataset is already fixed and new samples will not continue to be added, the maximum and minimum values of each feature will not change, so the normalization processing in this article uses the max-min mapping function, which is implemented with the help of the StandardScaler function under the sklearn library in Python, and the function form is as follows [32]: (3)

In the Eq (3): X^* is the mapped numeric value; x represents the value in the original data set; min is the minimum value of each feature; max is the maximum value for each feature.

4. Model introduction and model training

4.1 Extreme gradient boosting (XGBoost)

The extreme Gradient Boosting (XGBoost) model is based on the tree model, which is an improved model based on Gradient Boosting Decision Tree (GBDT). It is an addition formula composed of k base models proposed by Chen Tianqi, a scholar at the University of Washington. The basic idea is to use the new base model to fit the deviation of the previous model to continuously reduce the variation of the additive model [33]. By introducing regularization to the objective function and using the second derivative approximation, the XGBoost model effectively reduces the overfitting problem and improves the computational efficiency of the algorithm. Compared with other machine learning algorithms, the XGBoost algorithm is faster and more generalized. The prediction model formula is as follows: (4)

In the Eq (4): f_t is the kth base model; is the predicted value of the ith sample.

To avoid overfitting, the regularization term on a single base model is as follows: (5)

In the Eq (5): T is the leaf node of the tree; ω is the output value of x falling on a leaf node; γ and λ are non-negative coefficients; ω_j is the output value of the j_th node.

Therefore, the XGBoost loss function is defined as: (6)

The minimum value of the loss function is obtained using Taylor’s second-order expansion, and then the segmentation point with the highest score is searched by exact or approximate methods, and the next step is to segment and expand the leaf nodes [34].

4.2 Random Forest (RF)

The Random Forest (RF) model is based on the decision tree model under the bagging framework. The bagging with the decision tree as the primary model generates a decision tree after each bootstrap is put back into the sampling, there is no further intervention in developing these trees; random forest also performs bootstrap sampling, but the difference between it and bagging is that when generating each tree, each node variable is only generated in a few randomly selected variables, Therefore, not only the samples are random, but also the generation of each node variable is random; The advantage is that it can transform a limited number of weak classifiers into a strong classifier through linear combination, to improve the accuracy and robustness [35]. The problem studied in this paper is a regression problem. The random forest regression algorithm outputs the results of all decision trees and then takes the mean value. The regression equation is as follows: (7)

In the Eq (7): is the prediction result; hi is a single decision tree; θ_k is an independent distributed random variable that determines the growth process of a single decision tree; K is the number of decision trees.

Each decision tree in a random forest does not capture all features simultaneously, so each decision tree has uncertainty, so the model’s generalization ability is increased.

4.3 Support Vector Regression (SVR)

The Support Vector Machine (SVM) model is a data mining model based on statistical theory, and its purpose is to find an optimal hyperplane to separate two different classes. When Support Vector Regression is used for regression problems, the model is called Support Vector Regression(SVR); when dealing with SVR nonlinear problems, the kernel function maps nonlinear problems in a low-dimensional space to a high-dimensional feature space, and finds the optimal hyperplane in the high-dimensional space, and then calculate the distance between all training set samples and this plane, so that the distance is the smallest, to achieve regression prediction [36, 37]. In this paper, through multi-parameter verification, the kernel function used is RBF (Radial Basis Kernel Function), and the kernel function expression is as follows: (8)

In the Eq (8): σ is the width parameter of the RBF; exp is the exponential function with the natural constant e as the base; X-X_i is the distance between the selected center points.

High-dimensional feature space decision function: (9)

In the Eq (9): W is an adjustable weight vector; φ(x) is a nonlinear mapping function; b is a constant.

Ignoring the fitting error smaller than ε, the SVR model can be transformed into a constrained optimization problem as follows.

(10)

In the Eq (10): c is the penalty factor; are a pair of relaxation factors.

By introducing the Lagrange function, the optimization problem of Eq (10) is transformed into a dual form, and the SVR objective equation is obtained through the transformation and solution as shown in Eq (11): (11)

In the Eq (11): a* and a are dual variables; K(X_i,X_j) = φ(x_i)Tφ(x_j)。

4.4 Model parameter selection

The optimal parameters of the three models are found by grid search [38]. Grid search is a model parameter optimization technique that performs an exhaustive search for the specified parameter values. First, the Cartesian product of the given parameter values is performed to obtain a combination of a finite set of parameters. Each set of parameters is used to train the model, and select a set of parameters with the best performance as the optimal parameters. The model is written in Python language, and the grid search method uses the GridSearchCV function under the sklearn library. The optimal parameters of the three models are shown in Table 3.

Download:

Table 3. Optimal parameters of each model.

https://doi.org/10.1371/journal.pone.0279293.t003

5. Model prediction results and performance analysis

The trained model is applied to the fitting of the verification set, and the fitting graph of the actual value and the predicted value is obtained as shown in Fig 4. The fitting diagram of the XGBoost model is shown in a in Fig 4, the fitting diagram of the RF model is shown in b in Fig 4, and the fitting diagram of the SVR model is shown in c in Fig 4. It can be seen from the Fig 4(A) that the XGBoost model has higher prediction accuracy than the other two models.

Download:

Fig 4. Model fitting diagram.

(a) fitting diagram of XGBoost model true value and predicted value; (b) fitting diagram of RF model true value and predicted value; (c) fitting diagram of SVR model true value and predicted value.

https://doi.org/10.1371/journal.pone.0279293.g004

In order to evaluate the comprehensive performance of the model more objectively and combined with the fact that the problem in this study is a regression in nature, the Root Mean Square Error (RMSE), fitting coefficient (R²) and Mean Absolute Error (MAE) are adopted to evaluate indicators. The Root Mean Square Error (RMSE) is used to measure the degree of data change. The smaller the RMSE value is, the higher the accuracy is. The fitting coefficient R² is used to measure the fitting degree between the predicted value and the real value. The closer it is to 1, the better the model fitting effect is. MAE is the sum of the absolute value difference between the real value and the predicted value. The closer it is to 0, the better the model performance is. The evaluation indices of the three models are shown in Table 4. The XGBoost model has the highest R² value and the lowest RMSE and MAE value, so it performs best. At the same time, the RF model has the second highest R² value and the second lowest RMSE and MAE values, which are lower than the XGBoost model in performance, and the SVR model performs poorly. The performance of traditional SVR and RF models is lower than XGBoost models. The three parameter formulas are as follows [39]: (12) (13) (14)

Download:

Table 4. Performance index of each model.

https://doi.org/10.1371/journal.pone.0279293.t004

In the Eqs (12)–(14): n is the number of data; b_ti is the actual value of the t feature of the i_th sample; b_pi is the predicted value of the p feature of the ith sample, and are the average of the actual and predicted results, respectively.

The applied predictive models were assessed using the feasibility of correlation analysis. Scatter plot is one of the informative graphical presentations for examining the aptitude of prediction models [40, 41]. Based on the exhibited results in Fig 5, Wherein, Fig 5(A) shows the scatter plot of the prediction results of XGBoost model, Fig 5(B) shows the scatter plot of the prediction results of RF model, and Fig 5(C) shows the scatter plot of the prediction results of SVR model. Comparing the three figures, it is found that XGBoost model can well mine the relationship between input and output after training, and the correlation between the prediction results and experimental results is closer to 1 compared with RF model and SVR model.

Download:

Fig 5. Scatter plots demonstration between actual compressive strength and predicted values obtained from machine learning models.

https://doi.org/10.1371/journal.pone.0279293.g005

In the testing phase, a total of 206 samples were tested, and the relative error (RE) percentage was calculated for each sample, and the RE value can visually observe the stability of the prediction model and the error extremes, as shown in Fig 6, Fig 6(A) is the relative error percentage of the XGBoost model, Fig 6(B) is the relative error percentage of the RF model, Fig 6(C) is the relative error percentage of the SVR model. The XGBoost model is more stable, with the absolute value of the error percentage within 10%, while the absolute value of the error percentage of the RF and SVR models is within 27%; in terms of the error extremes, the absolute value of the XGBoost model is 8%, while the RF model reaches 26% and the SVR model reaches 27%, based on the RE% evaluation, the XGBoost model is more stable in prediction performance [42].

Download:

Fig 6. Relative error-index over the test phase sample for predicted values utilizing machine learning models.

https://doi.org/10.1371/journal.pone.0279293.g006

Fig 7 summarizes the general performances of the applied models in the form of the Taylor diagram [43]. This diagram simply expresses three main statistics including the correlation coefficient between the predicted values and measured data (as an angle in the polar plot), RMSE (as a radial distance from the observation point), and the ratio of the standard deviation of the predicted values (as a radial distance from the origin). As can be observed in Fig 7, the XGBoost molel is closer to the REF point in comparison with the other two predictive modelling approaches (RF, and SVR).

Download:

Fig 7. Taylor diagram presentation for the predicted values of compressive strength utilizing machine learning model.

https://doi.org/10.1371/journal.pone.0279293.g007

6. Experimental verification

In laboratory was configured with a 100×100×100mm concrete test block with the same characteristics as the dataset. The test block is placed in a constant temperature curing box and cured for 28 days at 18 degrees Celsius humidity of 95%, as shown in Fig 8. To prevent the occurrence of accidents such as damage and increase the number of samples to reduce the contingency, 10 groups of proportioning are prepared for this time, and 3 test blocks are made for each proportioning, with a total of 30 test blocks. The concrete compressive strength test was carried out on the test block with the ideal appearance among the three test blocks of each mixing ratio; the concrete mix ratio, predicted strength, and experimental strength was obtained, as shown in Table 5. Finally, the average error rate of 10 combinations of test blocks is calculated, and the average error rate is 3.5%, which is low, and the model’s accuracy is in line with the actual engineering needs [44]. The formula for the average error rate is as follows: (15)

Download:

Fig 8.

Curing of test block.

https://doi.org/10.1371/journal.pone.0279293.g008

Download:

Table 5. Actual and predicted compressive strength and error rate of each proportioned concrete test block.

https://doi.org/10.1371/journal.pone.0279293.t005

In the Eq (15): e is the average error rate; x* is the actual value; x is the predicted value; n is the total number of samples.

7. Conclusion

After the initial model training of the three machine learning models, the characteristic data are input into the model to predict the compressive strength of concrete. Then, the fitting curves of the actual values and predicted values of the three models are compared, and the data of performance indicators such as R², RMSE and MAE are compared. The results show that XGBoost has better performance in predicting concrete compressive strength. R², RMSE and MAE are 0.95, 0.05 and 0.15, respectively. R² is closer to 1, RMSE and MAE are closer to 0, and the error is smaller, more effective than SVR and RF models.
In the laboratory will insert fly ash and ground slag as admixture into concrete test block, using the hydraulic universal testing machine to validate its compressive strength, and use the trained XGBoost model to predict the compressive strength, the results show that the XGBoost models to predict the concrete compressive strength performance good, easy to model, computing speed is fast. The actual average error rate is only 3.5%, which provides an effective method for predicting the compressive strength of concrete with a known mix ratio in advance.
By using traditional solid waste fly ash and blast furnace slag as a concrete admixture to make concrete, the concrete finally obtained has high compressive strength and can be used stably for a long time, which meets the requirements of engineering strength. It has guiding significance for engineering applications and realizes the goal of recycling solid waste such as fly ash and slag.
The machine learning model is used to predict the compressive strength of concrete. The compressive strength data of concrete made using fly ash, and blast furnace slag as concrete mixing material can be obtained through fewer experiments. Compared with the previous experiments, concrete samples are prepared in large quantities to test their compressive strength. Predicting its compressive strength through a machine learning model saves experimental costs and resources.

8. Limitation and future research direction

This paper investigates the compressive strength of concrete with fly ash and slag as admixture materials. The mechanical properties, such as tensile and flexural strength, are not investigated. They need to be further explored to establish a complete predictive model for concrete mechanics of solid waste materials such as fly ash and slag.
In this paper, the compressive strength of concrete was investigated using the ratio of eight different materials as the input parameters of the prediction model. Still, many factors affect the compressive strength of concrete with solid waste materials such as fly ash and slag. There are constraints among the factors so that subsequent studies can consider as many different factors as possible as input parameters, such as aggregate particle size, each component of solid waste materials ratio, and other factors.

References

1. Mustapha R., Aroussi M. High-Performance Concrete Compressive Strength Prediction Based Weighted Support Vector Machines, International Journal of Engineering Research and Applications. 2017;07,68–75. https://doi.org/10.9790/9622-0701016875.
- View Article
- Google Scholar
2. Ly HaiBang, Nguyen ThuyAnh, and Pham Binh Thai. Investigation on factors affecting early strength of high-performance concrete by Gaussian Process Regression. PloS one. 2022;17.1. pmid:35085343
- View Article
- PubMed/NCBI
- Google Scholar
3. Mehta PK. Global concrete industry sustainability: tools for moving forward to cut carbon emissions. In: Concrete international, February 2009. p. 45–8.
4. Malhotra M. Reducing CO2 emissions: the role of fly ash and other supplementary cementitious materials. In: Concrete international, September 2006. p. 42–5.
5. Mohamed S. B. Tayeb. Effect of coarse aggregates and sand contents on workability and static stability of self-compacting concrete. Advances in Concrete Construction. 2019;7(2):97–105. https://doi.org/10.12989/acc.2019.7.2.097
- View Article
- Google Scholar
6. Srinivas K., Sankar L. Ponraj, Swamy C. Kumara. Experimental investigation on rapid strength gain by adding alccofine in high strength concrete. Materials Today: Proceedings. 2021;46.P1.
- View Article
- Google Scholar
7. Saber Ayad Zaki. Prediction and developing of shear strength of reinforced high strength concrete beams with and without steel fibers using multiple mathematical models. PloS one. 2022;17.3. pmid:35358237
- View Article
- PubMed/NCBI
- Google Scholar
8. Netinger I, Bjegovi D. Utilisation of steel slag as an aggregate in concrete. Materials & Structures. 2011;44(9):1565–1575. https://doi.org/10.1617/s11527-011-9719-8
- View Article
- Google Scholar
9. Khosravi K., Shahabi H., Pham B.T., Adamowski J., Shirzadi A., Pradhan B. A comparative assessment of flood susceptibility modeling using Multi-Criteria Decision-Making Analysis and Machine Learning Methods. Journal of Hydrology. 2019;573: 311–323. https://doi.org/10.1016/j.jhydrol.2019.03.073.
- View Article
- Google Scholar
10. Panagiotis G. Asteris, Vaseilios G. Mokos. Concrete compressive strength using artificial neural networks. Neural Computing and Applications.2019;32.15.
- View Article
- Google Scholar
11. Ly Hai-Bang. Development of deep neural network model to predict the compressive strength of rubber concrete. Construction and Building Materials. 2021;301.
- View Article
- Google Scholar
12. Ly Hai Bang, Nguyen May Huu, Pham Binh Thai. Metaheuristic optimization of Levenberg–Marquardt-based artificial neural network using particle swarm optimization for prediction of foamed concrete compressive strength. Neural Computing and Applications. 2021;33.24.
- View Article
- Google Scholar
13. Nguyen Thuy Anh, Ly Hai Bang, Tran Van Quan. Investigation of ANN Architecture for Predicting Load-Carrying Capacity of Castellated Steel Beams. COMPLEXITY. 2021;2021.
- View Article
- Google Scholar
14. Nguyen Thuy Anh. On the Training Algorithms for Artificial Neural Network in Predicting the Shear Strength of Deep Beams. COMPLEXITY. 2021;2021.
- View Article
- Google Scholar
15. Yeh I.-C. Modeling of strength of high-performance concrete using artificial neural networks, Cem. Concr. Res. (1998) 1797–1808.
- View Article
- Google Scholar
16. Yeh I.-C. Modeling concrete strength with augment-neuron networks, Journal of Materials in Civil Engineering. 10, n (1998) 263–268.
- View Article
- Google Scholar
17. Yeh I.-C. Design of high-performance concrete mixture using neural networks and nonlinear programming, Journal of Computing in Civil Engineering. 13, n (1999) 36–42.
- View Article
- Google Scholar
18. Fan Z. R. Chiong. A fuzzy weighted relative error support vector machine for reverse prediction of concrete components. Computers & Structures. 2020;230:(106171–1061783). https://doi.org/10.1016/j.compstruc.2019.106171
- View Article
- Google Scholar
19. Behnood A., Golafshani E. M. Machine learning study of the mechanical properties of concretes containing waste foundry sand. Construction and Building Materials. 2020;243(4), 118152. https://doi.org/10.1016/j.conbuildmat.2020.118152
- View Article
- Google Scholar
20. Deshpande N., Londhe S. Modeling compressive strength of recycled aggregate concrete by Artificial Neural Network, Model Tree and Non-linear Regression. International Journal of Sustainable Built Environment. 2014;3(2): 187–198. https://doi.org/10.1016/j.ijsbe.2014.12.002
- View Article
- Google Scholar
21. Saridemir M. Empirical modeling of flexural and splitting tensile strengths of concrete containing fly ash by GEP. Computers & Concrete.2019;17(4): 489–498. https://doi.org/10.12989/cac.2016.17.4.489
- View Article
- Google Scholar
22. Cho C. G., Hotta H. A study on compressive strength of concrete in flexural regions of reinforced concrete beams using finite element analysis. Structural Engineering & Mechanics. 2002;13(3): 313–328. https://doi.org/10.12989/sem.2002.13.3.313
- View Article
- Google Scholar
23. Xue X. Application of a support vector machine for prediction of piping and internal stability of soils. Geomechanics and Engineering.2019; 5(18): 493–502. https://doi.org/10.12989/gae.2019.18.5.493
- View Article
- Google Scholar
24. Rayed Alyousef H. A. A. M. A model to develop the porosity of concrete as important mechanical property. Smart Structures and Systems.2020;2(26): 147–156. https://doi.org/10.12989/sss.2020.26.2.147
- View Article
- Google Scholar
25. Mohammed Sonebi S. G. A. C. Modelling fresh properties of self-compacting concrete using neural network technique. Computers and Concrete. 2016;4(18): 903–921. https://doi.org/10.12989/cac.2016.18.4.903
- View Article
- Google Scholar
26. A, I. C. Y. Modeling of strength of high-performance concrete using artificial neural networks. Cement & Concrete Research.1998; 28(12): 1797–1808. https://doi.org/10.1016/S0008-8846(98)00165-3
- View Article
- Google Scholar
27. Kang Jiaxu. (2022): Concrete_Data. figshare. Dataset. https://doi.org/10.6084/m9.figshare.21493236.v1
- View Article
- Google Scholar
28. Kim D. and Pham K. Determination of effective parameters on surface settlement during shield TBM. Geomechanics and Engineering.2020;21(2): 153–164. https://doi.org/10.12989/gae.2020.21.2.153
- View Article
- Google Scholar
29. Adler J., Parmryd I. Quantifying colocalization by correlation: The Pearson correlation coefficient is superior to the Mander’s overlap coefficient. Cytometry Part A. 2010; 77a(8): 733–742. https://doi.org/10.1002/cyto.a.20896
- View Article
- Google Scholar
30. Hauke J., Kossowski T. COMPARISON OF VALUES OF PEARSON’S AND SPEARMAN’S CORRELATION COEFFICIENTS ON THE SAME SETS OF DATA. Quaestiones Geographicae.2011;30(2): 87–93. https://doi.org/10.2478/v10117-011-0021-1
- View Article
- Google Scholar
31. Fieller E. C. E. S. Pearson. Tests for rank correlation coefficients. II. Biometrika.1961;48(1–2): 29–40. https://doi.org/10.2307/2333127
- View Article
- Google Scholar
32. Mahdi Shariati M. S. M. P. Application of Extreme Learning Machine (ELM) and Genetic Programming (GP) to design steel-concrete composite floor systems at elevated temperatures. Steel and Composite Structures. 2019;33(3): 319–332. https://doi.org/10.12989/scs.2019.33.3.319
- View Article
- Google Scholar
33. Hao D, Xin X. Gaofen-3 PolSAR Image Classification via XGBoost and Polarimetric Spatial Information.Sensors. 2018;18(2): 611–. pmid:29462962
- View Article
- PubMed/NCBI
- Google Scholar
34. Sukhpreet D. N. Abdullah Al. Effective Intrusion Detection System Using XGBoost. Information. 2018;9(7): 149–. https://doi.org/10.3390/info9070149
- View Article
- Google Scholar
35. Arif M. Classification of cardiotocograms using random forest classifier and selection of important features from cardiotocogram signal. Biomaterials and Biomedical Engineering. 2015;2(3):173–183. https://doi.org/10.12989/bme.2015.2.3.173
- View Article
- Google Scholar
36. Kuang Yi Lin , T. K. L. Y.Real-time seismic structural response prediction system based on support vector machine. Earthquakes and Structures. 2020;18(2): 163–170. https://doi.org/10.12989/eas.2020.18.2.163
- View Article
- Google Scholar
37. Shi W. N. Wang. Quantifying the Independent Influences of Land Cover and Humidity on Microscale Urban Air Temperature Variation in Hot Summer: Methods of Path Analysis and Genetic SVR. Atmosphere. 2020;11(12): 1377. https://doi.org/10.3390/atmos11121377
- View Article
- Google Scholar
38. Belete D. M, Huchaiah M. D. Grid search in hyperparameter optimization of machine learning models for prediction of HIV/AIDS test results. International Journal of Computers and Applications. 2021;1: 1–12. https://doi.org/10.1080/1206212X.2021.1974663
- View Article
- Google Scholar
39. Nitisha Sharma A. U. M. S. Comparison of machine learning algorithms to evaluate strength of concrete with marble powder. Advances in Materials Research. 2022;11(1):75–90. https://doi.org/10.12989/amr.2022.11.1.075
- View Article
- Google Scholar
40. Ahmad S, Masoud. Mussa H, M.S.A, M.H, et al. Development of Advanced Computer Aid Model for Shear Strength of Concrete Slender Beam Prediction. Applied sciences, 2020;(10):3811. https://doi.org/10.3390/app10113811
- View Article
- Google Scholar
41. Zounemat-Kermani M., Alizamir M., Yaseen Z.M., et al. Concrete corrosion in wastewater systems: Prediction and sensitivity analysis using advanced extreme learning machine. Front. Struct. Civ. Eng. 2021; (15), 444–460. https://doi.org/10.1007/s11709-021-0697-9
- View Article
- Google Scholar
42. Zhang G., Ali Z.H., Aldlemy M.S., et al. Reinforced concrete deep beam shear strength capacity modelling using an integrative bio-inspired algorithm with an artificial intelligence model. Engineering with Computers 38 (Suppl 1), 2022; 15–28. https://doi.org/10.1007/s00366-020-01137-1
- View Article
- Google Scholar
43. Sharafati A., Naderpour H., Salih S.Q., et al. Simulation of foamed concrete compressive strength prediction using adaptive neuro-fuzzy inference system optimized by nature-inspired algorithms. Front. Struct. Civ. Eng. 2021; (15), 61–79. https://doi.org/10.1007/s11709-020-0684-6
- View Article
- Google Scholar
44. Yang Y, Zhan B., et al. Nondestructive assessment of the compressive strength of concrete with high volume slag. Materials Characterization. 2020;162: 110223. https://doi.org/10.1016/j.matchar.2020.110223
- View Article
- Google Scholar

[ref1] 1. Mustapha R., Aroussi M. High-Performance Concrete Compressive Strength Prediction Based Weighted Support Vector Machines, International Journal of Engineering Research and Applications. 2017;07,68–75. https://doi.org/10.9790/9622-0701016875.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Ly HaiBang, Nguyen ThuyAnh, and Pham Binh Thai. Investigation on factors affecting early strength of high-performance concrete by Gaussian Process Regression. PloS one. 2022;17.1. pmid:35085343
View Article
PubMed/NCBI
Google Scholar

[5] View Article

[6] PubMed/NCBI

[7] Google Scholar

[ref3] 3. Mehta PK. Global concrete industry sustainability: tools for moving forward to cut carbon emissions. In: Concrete international, February 2009. p. 45–8.

[ref4] 4. Malhotra M. Reducing CO2 emissions: the role of fly ash and other supplementary cementitious materials. In: Concrete international, September 2006. p. 42–5.

[ref5] 5. Mohamed S. B. Tayeb. Effect of coarse aggregates and sand contents on workability and static stability of self-compacting concrete. Advances in Concrete Construction. 2019;7(2):97–105. https://doi.org/10.12989/acc.2019.7.2.097
View Article
Google Scholar

[11] View Article

[12] Google Scholar

[ref6] 6. Srinivas K., Sankar L. Ponraj, Swamy C. Kumara. Experimental investigation on rapid strength gain by adding alccofine in high strength concrete. Materials Today: Proceedings. 2021;46.P1.
View Article
Google Scholar

[14] View Article

[15] Google Scholar

[ref7] 7. Saber Ayad Zaki. Prediction and developing of shear strength of reinforced high strength concrete beams with and without steel fibers using multiple mathematical models. PloS one. 2022;17.3. pmid:35358237
View Article
PubMed/NCBI
Google Scholar

[17] View Article

[18] PubMed/NCBI

[19] Google Scholar

[ref8] 8. Netinger I, Bjegovi D. Utilisation of steel slag as an aggregate in concrete. Materials & Structures. 2011;44(9):1565–1575. https://doi.org/10.1617/s11527-011-9719-8
View Article
Google Scholar

[21] View Article

[22] Google Scholar

[ref9] 9. Khosravi K., Shahabi H., Pham B.T., Adamowski J., Shirzadi A., Pradhan B. A comparative assessment of flood susceptibility modeling using Multi-Criteria Decision-Making Analysis and Machine Learning Methods. Journal of Hydrology. 2019;573: 311–323. https://doi.org/10.1016/j.jhydrol.2019.03.073.
View Article
Google Scholar

[24] View Article

[25] Google Scholar

[ref10] 10. Panagiotis G. Asteris, Vaseilios G. Mokos. Concrete compressive strength using artificial neural networks. Neural Computing and Applications.2019;32.15.
View Article
Google Scholar

[27] View Article

[28] Google Scholar

[ref11] 11. Ly Hai-Bang. Development of deep neural network model to predict the compressive strength of rubber concrete. Construction and Building Materials. 2021;301.
View Article
Google Scholar

[30] View Article

[31] Google Scholar

[ref12] 12. Ly Hai Bang, Nguyen May Huu, Pham Binh Thai. Metaheuristic optimization of Levenberg–Marquardt-based artificial neural network using particle swarm optimization for prediction of foamed concrete compressive strength. Neural Computing and Applications. 2021;33.24.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref13] 13. Nguyen Thuy Anh, Ly Hai Bang, Tran Van Quan. Investigation of ANN Architecture for Predicting Load-Carrying Capacity of Castellated Steel Beams. COMPLEXITY. 2021;2021.
View Article
Google Scholar

[36] View Article

[37] Google Scholar

[ref14] 14. Nguyen Thuy Anh. On the Training Algorithms for Artificial Neural Network in Predicting the Shear Strength of Deep Beams. COMPLEXITY. 2021;2021.
View Article
Google Scholar

[39] View Article

[40] Google Scholar

[ref15] 15. Yeh I.-C. Modeling of strength of high-performance concrete using artificial neural networks, Cem. Concr. Res. (1998) 1797–1808.
View Article
Google Scholar

[42] View Article

[43] Google Scholar

[ref16] 16. Yeh I.-C. Modeling concrete strength with augment-neuron networks, Journal of Materials in Civil Engineering. 10, n (1998) 263–268.
View Article
Google Scholar

[45] View Article

[46] Google Scholar

[ref17] 17. Yeh I.-C. Design of high-performance concrete mixture using neural networks and nonlinear programming, Journal of Computing in Civil Engineering. 13, n (1999) 36–42.
View Article
Google Scholar

[48] View Article

[49] Google Scholar

[ref18] 18. Fan Z. R. Chiong. A fuzzy weighted relative error support vector machine for reverse prediction of concrete components. Computers & Structures. 2020;230:(106171–1061783). https://doi.org/10.1016/j.compstruc.2019.106171
View Article
Google Scholar

[51] View Article

[52] Google Scholar

[ref19] 19. Behnood A., Golafshani E. M. Machine learning study of the mechanical properties of concretes containing waste foundry sand. Construction and Building Materials. 2020;243(4), 118152. https://doi.org/10.1016/j.conbuildmat.2020.118152
View Article
Google Scholar

[54] View Article

[55] Google Scholar

[ref20] 20. Deshpande N., Londhe S. Modeling compressive strength of recycled aggregate concrete by Artificial Neural Network, Model Tree and Non-linear Regression. International Journal of Sustainable Built Environment. 2014;3(2): 187–198. https://doi.org/10.1016/j.ijsbe.2014.12.002
View Article
Google Scholar

[57] View Article

[58] Google Scholar

[ref21] 21. Saridemir M. Empirical modeling of flexural and splitting tensile strengths of concrete containing fly ash by GEP. Computers & Concrete.2019;17(4): 489–498. https://doi.org/10.12989/cac.2016.17.4.489
View Article
Google Scholar

[60] View Article

[61] Google Scholar

[ref22] 22. Cho C. G., Hotta H. A study on compressive strength of concrete in flexural regions of reinforced concrete beams using finite element analysis. Structural Engineering & Mechanics. 2002;13(3): 313–328. https://doi.org/10.12989/sem.2002.13.3.313
View Article
Google Scholar

[63] View Article

[64] Google Scholar

[ref23] 23. Xue X. Application of a support vector machine for prediction of piping and internal stability of soils. Geomechanics and Engineering.2019; 5(18): 493–502. https://doi.org/10.12989/gae.2019.18.5.493
View Article
Google Scholar

[66] View Article

[67] Google Scholar

[ref24] 24. Rayed Alyousef H. A. A. M. A model to develop the porosity of concrete as important mechanical property. Smart Structures and Systems.2020;2(26): 147–156. https://doi.org/10.12989/sss.2020.26.2.147
View Article
Google Scholar

[69] View Article

[70] Google Scholar

[ref25] 25. Mohammed Sonebi S. G. A. C. Modelling fresh properties of self-compacting concrete using neural network technique. Computers and Concrete. 2016;4(18): 903–921. https://doi.org/10.12989/cac.2016.18.4.903
View Article
Google Scholar

[72] View Article

[73] Google Scholar

[ref26] 26. A, I. C. Y. Modeling of strength of high-performance concrete using artificial neural networks. Cement & Concrete Research.1998; 28(12): 1797–1808. https://doi.org/10.1016/S0008-8846(98)00165-3
View Article
Google Scholar

[75] View Article

[76] Google Scholar

[ref27] 27. Kang Jiaxu. (2022): Concrete_Data. figshare. Dataset. https://doi.org/10.6084/m9.figshare.21493236.v1
View Article
Google Scholar

[78] View Article

[79] Google Scholar

[ref28] 28. Kim D. and Pham K. Determination of effective parameters on surface settlement during shield TBM. Geomechanics and Engineering.2020;21(2): 153–164. https://doi.org/10.12989/gae.2020.21.2.153
View Article
Google Scholar

[81] View Article

[82] Google Scholar

[ref29] 29. Adler J., Parmryd I. Quantifying colocalization by correlation: The Pearson correlation coefficient is superior to the Mander’s overlap coefficient. Cytometry Part A. 2010; 77a(8): 733–742. https://doi.org/10.1002/cyto.a.20896
View Article
Google Scholar

[84] View Article

[85] Google Scholar

[ref30] 30. Hauke J., Kossowski T. COMPARISON OF VALUES OF PEARSON’S AND SPEARMAN’S CORRELATION COEFFICIENTS ON THE SAME SETS OF DATA. Quaestiones Geographicae.2011;30(2): 87–93. https://doi.org/10.2478/v10117-011-0021-1
View Article
Google Scholar

[87] View Article

[88] Google Scholar

[ref31] 31. Fieller E. C. E. S. Pearson. Tests for rank correlation coefficients. II. Biometrika.1961;48(1–2): 29–40. https://doi.org/10.2307/2333127
View Article
Google Scholar

[90] View Article

[91] Google Scholar

[ref32] 32. Mahdi Shariati M. S. M. P. Application of Extreme Learning Machine (ELM) and Genetic Programming (GP) to design steel-concrete composite floor systems at elevated temperatures. Steel and Composite Structures. 2019;33(3): 319–332. https://doi.org/10.12989/scs.2019.33.3.319
View Article
Google Scholar

[93] View Article

[94] Google Scholar

[ref33] 33. Hao D, Xin X. Gaofen-3 PolSAR Image Classification via XGBoost and Polarimetric Spatial Information.Sensors. 2018;18(2): 611–. pmid:29462962
View Article
PubMed/NCBI
Google Scholar

[96] View Article

[97] PubMed/NCBI

[98] Google Scholar

[ref34] 34. Sukhpreet D. N. Abdullah Al. Effective Intrusion Detection System Using XGBoost. Information. 2018;9(7): 149–. https://doi.org/10.3390/info9070149
View Article
Google Scholar

[100] View Article

[101] Google Scholar

[ref35] 35. Arif M. Classification of cardiotocograms using random forest classifier and selection of important features from cardiotocogram signal. Biomaterials and Biomedical Engineering. 2015;2(3):173–183. https://doi.org/10.12989/bme.2015.2.3.173
View Article
Google Scholar

[103] View Article

[104] Google Scholar

[ref36] 36. Kuang Yi Lin , T. K. L. Y.Real-time seismic structural response prediction system based on support vector machine. Earthquakes and Structures. 2020;18(2): 163–170. https://doi.org/10.12989/eas.2020.18.2.163
View Article
Google Scholar

[106] View Article

[107] Google Scholar

[ref37] 37. Shi W. N. Wang. Quantifying the Independent Influences of Land Cover and Humidity on Microscale Urban Air Temperature Variation in Hot Summer: Methods of Path Analysis and Genetic SVR. Atmosphere. 2020;11(12): 1377. https://doi.org/10.3390/atmos11121377
View Article
Google Scholar

[109] View Article

[110] Google Scholar

[ref38] 38. Belete D. M, Huchaiah M. D. Grid search in hyperparameter optimization of machine learning models for prediction of HIV/AIDS test results. International Journal of Computers and Applications. 2021;1: 1–12. https://doi.org/10.1080/1206212X.2021.1974663
View Article
Google Scholar

[112] View Article

[113] Google Scholar

[ref39] 39. Nitisha Sharma A. U. M. S. Comparison of machine learning algorithms to evaluate strength of concrete with marble powder. Advances in Materials Research. 2022;11(1):75–90. https://doi.org/10.12989/amr.2022.11.1.075
View Article
Google Scholar

[115] View Article

[116] Google Scholar

[ref40] 40. Ahmad S, Masoud. Mussa H, M.S.A, M.H, et al. Development of Advanced Computer Aid Model for Shear Strength of Concrete Slender Beam Prediction. Applied sciences, 2020;(10):3811. https://doi.org/10.3390/app10113811
View Article
Google Scholar

[118] View Article

[119] Google Scholar

[ref41] 41. Zounemat-Kermani M., Alizamir M., Yaseen Z.M., et al. Concrete corrosion in wastewater systems: Prediction and sensitivity analysis using advanced extreme learning machine. Front. Struct. Civ. Eng. 2021; (15), 444–460. https://doi.org/10.1007/s11709-021-0697-9
View Article
Google Scholar

[121] View Article

[122] Google Scholar

[ref42] 42. Zhang G., Ali Z.H., Aldlemy M.S., et al. Reinforced concrete deep beam shear strength capacity modelling using an integrative bio-inspired algorithm with an artificial intelligence model. Engineering with Computers 38 (Suppl 1), 2022; 15–28. https://doi.org/10.1007/s00366-020-01137-1
View Article
Google Scholar

[124] View Article

[125] Google Scholar

[ref43] 43. Sharafati A., Naderpour H., Salih S.Q., et al. Simulation of foamed concrete compressive strength prediction using adaptive neuro-fuzzy inference system optimized by nature-inspired algorithms. Front. Struct. Civ. Eng. 2021; (15), 61–79. https://doi.org/10.1007/s11709-020-0684-6
View Article
Google Scholar

[127] View Article

[128] Google Scholar

[ref44] 44. Yang Y, Zhan B., et al. Nondestructive assessment of the compressive strength of concrete with high volume slag. Materials Characterization. 2020;162: 110223. https://doi.org/10.1016/j.matchar.2020.110223
View Article
Google Scholar

[130] View Article

[131] Google Scholar

Figures

Abstract

1. Introduction

2. Research methods and steps

3. Data analysis

3.1 Sample data acquisition and preprocessing

3.2 Correlation analysis of sample data

3.3 Normalization processing

4. Model introduction and model training

4.1 Extreme gradient boosting (XGBoost)

4.2 Random Forest (RF)

4.3 Support Vector Regression (SVR)

4.4 Model parameter selection

5. Model prediction results and performance analysis

6. Experimental verification

7. Conclusion

8. Limitation and future research direction

References