# Application of Artificial Intelligence Methods for Predicting the Compressive Strength of Self-Compacting Concrete with Class F Fly Ash

^{1}

^{2}

^{3}

^{*}

## Abstract

**:**

## 1. Introduction

^{®}were used as accuracy criteria. In their paper, Asteris et al. [8] investigated the application of ANN in predicting the CS of SCC with the addition of fly ash after 28 days. The basis for the development of the model was tests of 169 samples collected from the published literature. Douma et al. [9] investigated the application of ANN to model the properties of SCC with the addition of fly ash. The prediction of fresh concrete properties and CS after 28 days was analyzed. Models with a total of 6 constituents as input variables were analyzed, and the base consisted of 114 examined samples. Models were evaluated through MSE, coefficient of determination, and MAPE criteria. The research recommended an ANN model with one hidden layer of 17 neurons as optimal. Asteris and Kolovos [10] worked on the application of a surrogate model to assess the CS of SCC after 28 days. Eleven different constituents of SCC were considered, one of which was fly ash. Different architectures of ANNs trained on a set of data from 205 examined samples were analyzed, and their accuracy was evaluated through the correlation coefficient R. They recommended using ANN as an optimal model. Saha et al. [11] researched the application of the support vector machines (SVM) model with different kernel functions in predicting the properties of fresh and hardened concrete with the addition of fly ash. The models were evaluated via a correlation coefficient. A model with an exponential radial basis function (ERBF) was recommended as the optimal model. Research related to the examination of the properties of fresh and hardened high-volume fly ash concrete was conducted by Azimi-Pour and others [12]. The application of SVM with different linear and nonlinear kernel functions was tested, based on the tested samples in a fresh and hardened state. The determination coefficient, RMSE, and MAPE were used as accuracy criteria. A model with an RBF kernel function was recommended as a model with higher accuracy compared to others. In their study of CS prediction modeling lightweight self-compacting concrete with the addition of fly ash, Zhang et al. [13] analyzed hybrid procedures in which they combined the beetle antennae search algorithm (BAS) with the random forest (RF) algorithm. The BAS algorithm was used to optimize the hyperparameters of the RF model. Song et al. [14] worked on the application of regression trees (RT), ANNs, genetic engineering programming, and boosting regressor models in the development of models for predicting the CS of SCC. The accuracy of the model was assessed using cross-validation, and the criteria used were the coefficient of determination (${\mathrm{R}}^{2}$), root mean error (RME), and root mean squared error (RMSE). Research recommended the use of ensemble algorithms in terms of accuracy. Hadzima-Nyarko et al. [15] investigated the application of SCC with the use of rubber and fly ash additives. The paper analyzed the application of different GPR models. This study showed that Gaussian process regression (GPR) modeling is an appropriate method for predicting the CS of SCC with recycled tire rubber particles and fly ash. Their results were further confirmed by scanning electron microscopy (SEM) images. Kovacevic et al. [16] conducted a similar study to create a model for predicting the CS of lightweight concrete with the addition of rubber and fly ash. The research concluded that GPR models are the optimal choice in this case. The combination of ANN models where network parameters are optimized using the firefly optimization algorithm in the prediction of CS samples of different ages was considered by Pazouki et al. [17]. Farooq et al. [18] performed research on determining a suitable model for predicting the CS of concrete modified with fly ash after 28 days. Models of ANNs, support vector machines, and gene expression programming (GEP) models were tested. The GEP model was proposed as the optimal model. In their study, de-Prado-Gil et al. dealt with the application of the ensemble methods: random forest (RF), K-nearest neighbor (KNN), extremely randomized trees (ERT), extreme gradient boosting (XGB), gradient boosting (GB), light gradient boosting machine (LGBM), category boosting (CB) and the generalized additive models (GAMs), and for the development of the models, 515 samples were collected. The results indicated that the RF models have a strong potential to predict the CS of SCC with recycled aggregates [19].

## 2. Methods

#### 2.1. Multi-Gene Genetic Programming (MGGP)

#### 2.2. Regression Tree Ensembles

#### 2.3. Support Vector Regression (SVR)

_{i}denotes the corresponding answers to the values of the input vectors, their values make up the following dataset $\left\{({x}_{1},{y}_{1}),({x}_{2},{y}_{2}),\dots ,({x}_{n},{y}_{n}),\right\}\in {R}^{m}\times R$.

#### 2.4. Gaussian Process Regression (GPR)

**model parameters**.

#### 2.5. Artificial Neural Networks (ANNs)

- A set of connections (synapses) where each connection has a certain weight;
- A sum function where the input signals are collected;
- Activation function, which limits the output of neurons.

## 3. Dataset

**randperm**function of the Matlab program, and then randomly 80% of the samples for model training and 20% for model testing. This 80:20 division of data is a standard procedure in machine learning (Table 2). All analyzed models were trained on an identical training set, while the accuracy of the model was assessed using the criteria root mean square error (RMSE), mean absolute error (MAE), Pearson’s linear correlation coefficient ©, and mean absolute percentage error (MAPE) on the identical test set.

## 4. Results

- Number of generated trees B;
- The minimum value of a leaf size (min leaf size) that represents the minimum amount of data assigned to a leaf within the tree;
- Number of randomly selected variables from the whole set of variables on which tree splitting will be performed (RF method only).

**Figure 14.**Comparison of different accuracy criteria for RF model as a function of the number of randomly selected splitting variables and minimum leaf size: (

**a**) RMSE, (

**b**) MAE, (

**c**) MAPE, and (

**d**) R.

- Number of generated trees B;
- Reduction parameter λ (learning rate);
- Number of splits of the tree d.

**Figure 15.**Dependence of the MSE value on the reduction parameter λ and the number of trees (base models) in the boosted trees model: (

**a**) the maximum number of splits is limited to 16 and (

**b**) the maximum number of splits is limited to 64.

- When applying the linear kernel in this study, the values for parameter $C$ in exponential order in the range $C=\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},\dots ,{2}^{9},{2}^{10}\right]$ were examined as well as the values for the parameter $\epsilon $ in the range $\epsilon =\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},{2}^{1}\right]$;
- For the RBF kernel, the values for parameter $C$ in the exponential order in the range $C=\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},\dots ,{2}^{9},{2}^{10}\right]$, the values of γ in the range $\gamma =\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},\dots ,{2}^{9},{2}^{10}\right]$, as well as the values for the parameter $\epsilon $ in the range $\epsilon =\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},{2}^{1}\right]$ were examined;
- For the sigmoid kernel, the values for the parameter $C$ in the exponential order in the range $C=\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},\dots ,{2}^{9},{2}^{10}\right]$, $\gamma $ in the range $\gamma =\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},\dots ,{2}^{9},{2}^{10}\right]$, as well as the values for the parameter $\epsilon $ in the range $\epsilon =\left[{2}^{-10},{2}^{-9},\dots ,{2}^{0},{2}^{1}\right]$ were examined. A value of zero was adopted for the value of the parameter $r$ ($r=0$) in accordance with the recommendation [56];
- All possible combinations of parameters within the above ranges were analyzed.

- Covariance functions that have one length scale parameter for all input variables (exponential, quadratic-exponential, Matérn 3/2, Matérn 5/2, rational square);
- Covariance functions that apply different length scale parameters to input variables (ARD covariance functions).

## 5. Conclusions

## Supplementary Materials

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## Appendix A

## References

- ASTM C618-05; Standard Specification for Fly Ash and Raw or Calcined Natural Pozzolan for Use as Mineral Admixture in Portland Cement Concrete. Annual Book of ASTM Standards. ASTM: West Conshohocken, PA, USA, 2005.
- Environmental Protection Agency (EPA); Federal Highway Administration (FHWA). Using Coal Ash in Highway Construction—A Guide to Benefits and Impacts; Report nr EPA-530-K-002:ID: 151; EPA: Washington, DC, USA, 2005.
- Federal Highway Administration (FHWA); American Coal Ash Association (ACAA). Fly Ash Facts for Highway Engineers; Report nr FHWA-IF-03-019; Federal Highway Administration (FHWA): Washington, DC, USA, 2003.
- Martínez-García, R.; Guerra-Romero, M.I.; Morán-del Pozo, J.M.; Brito, J.d.; Juan-Valdés, A. Recycling Aggregates for Self-Compacting Concrete Production: A Feasible Option. Materials
**2020**, 13, 868. [Google Scholar] [CrossRef] [PubMed] - Martínez-García, R.; Jagadesh, P.; Fraile-Fernández, F.J.; Morán-del Pozo, J.M.; Juan-Valdés, A. Influence of Design Parameters on Fresh Properties of Self-Compacting Concrete with Recycled Aggregate—A Review. Materials
**2020**, 13, 5749. [Google Scholar] [CrossRef] [PubMed] - Ahmad, J.; Zaid, O.; Aslam, F.; Martínez-García, R.; Alharthi, Y.M.; Hechmi EI Ouni, M.; Faisal Tufail, R.; Sharaky, I.A. Mechanical properties and durability assessment of nylon fiber reinforced self-compacting concrete. J. Eng. Fibers Fabr.
**2021**, 16, 1–13. [Google Scholar] [CrossRef] - Siddique, R.; Aggarwal, P.; Aggarwal, Y. Prediction of compressive strength of self-compacting concrete containing bottom ash using artificial neural networks. Adv. Eng. Softw.
**2011**, 42, 780–786. [Google Scholar] [CrossRef] - Asteris, P.G.; Kolovos, K.G.; Douvika, M.G.; Roinos, K. Prediction of self-compacting concrete strength using artificial neural networks. Eur. J. Environ. Civ. Eng.
**2016**, 20 (Suppl. S1), 102–122. [Google Scholar] [CrossRef] - Belalia Douma, O.; Boukhatem, B.; Ghrici, M. Prediction of properties of self-compacting concrete containing fly ash using artificial neural network. Neural Comput. Appl.
**2017**, 28, 707–718. [Google Scholar] [CrossRef] - Asteris, P.G.; Kolovos, K.G. Self-compacting concrete strength prediction using surrogate models. Neural Comput. Appl.
**2019**, 31, 409–424. [Google Scholar] [CrossRef] - Saha, P.; Debnath, P.; Thomas, P. Prediction of fresh and hardened properties of self-compacting concrete using support vector regression approach. Neural. Comput. Appl.
**2020**, 32, 7995–8010. [Google Scholar] [CrossRef] - Azimi-Pour, M.; Eskandari-Naddaf, H.; Pakzad, A. Linear and non-linear SVM prediction for fresh properties and compressive strength of high volume fly ash self-compacting concrete. Constr. Build. Mater.
**2020**, 230, 117021. [Google Scholar] [CrossRef] - Zhang, J.; Ma, G.; Huang, Y.; Sun, J.; Aslani, F.; Nener, B. Modelling uniaxial compressive strength of lightweight self-compacting concrete using random forest regression. Constr. Build. Mater.
**2019**, 210, 713–719. [Google Scholar] [CrossRef] - Song, H.; Ahmad, A.; Farooq, F.; Ostrowski, K.A.; Maślak, M.; Czarnecki, S.; Aslam, F. Predicting the compressive strength of concrete with fly ash admixture using machine learning algorithms. Constr. Build. Mater.
**2021**, 308, 125021. [Google Scholar] [CrossRef] - Hadzima-Nyarko, M.; Nyarko, E.K.; Djikanovic, D.; Brankovic, G. Microstructural and mechanical characteristics of self-compacting concrete with waste rubber. Struct. Eng. Mech.
**2021**, 78, 175–186. [Google Scholar] [CrossRef] - Kovačević, M.; Lozančić, S.; Nyarko, E.K.; Hadzima-Nyarko, M. Modeling of Compressive Strength of Self-Compacting Rubberized Concrete Using Machine Learning. Materials
**2021**, 14, 4346. [Google Scholar] [CrossRef] - Pazouki, G.; Golafshani, E.M.; Behnood, A. Predicting the compressive strength of self-compacting concrete containing Class F fly ash using metaheuristic radial basis function neural network. Struct. Concr.
**2022**, 23, 1191–1213. [Google Scholar] [CrossRef] - Farooq, F.; Czarnecki, S.; Niewiadomski, P.; Aslam, F.; Alabduljabbar, H.; Ostrowski, K.A.; Śliwa-Wieczorek, K.; Nowobilski, T.; Malazdrewicz, S. A Comparative Study for the Prediction of the Compressive Strength of Self-Compacting Concrete Modified with Fly Ash. Materials
**2021**, 14, 4934. [Google Scholar] [CrossRef] - De-Prado-Gil, J.; Palencia, C.; Silva-Monteiro, N.; Martínez-García, R. To predict the compressive strength of self compacting concrete with recycled aggregates utilizing ensemble machine learning models. Case Stud. Constr. Mater.
**2022**, 16, e01046. [Google Scholar] [CrossRef] - Searson, D.P.; Leahy, D.E.; Willis, M.J. GPTIPS: An Open Source Genetic Programming Toolbox for Multigene Symbolic Regression. In Proceeding of the International MultiConference of Engineers and Computer Scieintist Volume I, IMECS 2010, Hong Kong, China, 17–19 March 2010. [Google Scholar]
- Hastie, T.; Tibsirani, R.; Friedman, J. The Elements of Statistical Learning; Springer: Berlin/Heidelberg, Germany, 2009. [Google Scholar]
- Breiman, L.; Friedman, H.; Olsen, R.; Stone, C.J. Classification and Regression Trees; Chapman and Hall/CRC: Wadsworth, OH, USA, 1984. [Google Scholar]
- Breiman, L. Bagging predictors. Mach. Learn.
**1996**, 24, 123–140. [Google Scholar] [CrossRef] - Breiman, L. Random forests. Mach. Learn.
**2001**, 45, 5–32. [Google Scholar] [CrossRef] - Kovačević, M.; Ivanišević, N.; Petronijević, P.; Despotović, V. Construction cost estimation of reinforced and prestressed concrete bridges using machine learning. Građevinar
**2021**, 73, 1–13. [Google Scholar] [CrossRef] - Freund, Y.; Schapire, R.E. A decision-theoretic generalization of on-line learning and an application to boosting. J. Comput. Syst. Sci.
**1997**, 55, 119–139. [Google Scholar] [CrossRef] - Elith, J.; Leathwick, J.R.; Hastie, T. A working guide to boosted regression trees. J. Anim. Ecol.
**2008**, 77, 802–813. [Google Scholar] [CrossRef] - Friedman, J.H. Greedy function approximation: A gradient boosting machine. Ann. Stat.
**2001**, 29, 1189–1232. [Google Scholar] [CrossRef] - Vapnik, V. The Nature of Statistical Learning Theory; Springer: New York, NY, USA, 1995. [Google Scholar]
- Kecman, V. Learning and Soft Computing: Support Vector Machines, Neural Networks, and Fuzzy Logic Models; MIT Press: Cambridge, MA, USA, 2001. [Google Scholar]
- Smola, A.J.; Sholkopf, B. A tutorial on support vector regression. Stat. Comput.
**2004**, 14, 199–222. [Google Scholar] [CrossRef] - Rasmussen, C.E.; Williams, C.K. Gaussian Processes for Machine Learning; The MIT Press: Cambridge, MA, USA, 2006. [Google Scholar]
- Hagan, M.T.; Menhaj, M.B. Training Feedforward Networks with Marquardt Algorithm. IEEE Trans. Neural Netw.
**1994**, 5, 989–993. [Google Scholar] [CrossRef] [PubMed] - Khatib, J.M. Performance of self-compacting concrete containing fly ash. Construct. Build. Mater.
**2008**, 22, 1963–1971. [Google Scholar] [CrossRef] - Sonebi, M. Medium strength self-compacting concrete containing fly ash: Modelling using factorial experimental plans. Cem. Concr. Res.
**2004**, 34, 1199–1208. [Google Scholar] [CrossRef] - Bingöl, A.F.; Tohumcu, I. Effects of different curing regimes on the compressive strength properties of self compacting concrete incorporating fly ash and silica fume. Mater. Des.
**2013**, 51, 12–18. [Google Scholar] [CrossRef] - Bouzoubaâ, N.; Lachemi, M. Self-compacting concrete incorporating high volumes of class F fly ash: Preliminary results. Cem. Concr. Res.
**2001**, 31, 413–420. [Google Scholar] [CrossRef] - Mohamed, H.A. Effect of fly ash and silica fume on compressive strength of self-compacting concrete under different curing conditions. Ain. Shams. Eng. J.
**2011**, 2, 79–86. [Google Scholar] [CrossRef] - Sonebi, M.; Cevik, A. Prediction of fresh and hardened properties of self-consolidating concrete using neuro fuzzy approach. J. Mater. Civil. Eng.
**2009**, 21, 672–679. [Google Scholar] [CrossRef] - Bui, V.K.; Akkaya, Y.; Shah, S.P. Rheological model for self-consolidating concrete. ACI Mater. J.
**2002**, 99, 549–559. [Google Scholar] - Sonebi, M. Development of statistical models for mixture design of high-volume fly ash self-consolidating concrete. ACI Mater. J.
**2004**, 101, 339–346. [Google Scholar] [CrossRef] - Pathak, N.; Siddique, R. Properties of self-compacting-concrete containing fly ash subjected to elevated temperatures. Construct. Build. Mater.
**2012**, 30, 274–280. [Google Scholar] [CrossRef] - Pofale, A.D.; Deo, S.V. Comparative long term study of concrete mix design procedure for fine aggregate replacement with fly ash by minimum voids method and maximum density method. KSCE J. Civil. Eng.
**2010**, 14, 759–764. [Google Scholar] [CrossRef] - Nehdi, M.; Pardhan, M.; Koshowski, S. Durability of self-consolidating concrete incorporating high-volume replacement composite cements. Cem. Concr. Res.
**2004**, 34, 2103–2112. [Google Scholar] [CrossRef] - Sahmaran, M.; Lachemi, M.; Erdem, T.K.; Yücel, H.E. Use of spent foundry sand and fly ash for the development of green self-consolidating concrete. Mater. Struct. Construct.
**2011**, 44, 1193–1204. [Google Scholar] [CrossRef] - Ulucan, Z.Ç.; Türk, K.; Karatas, M. Effect of mineral admixtures on the correlation between ultrasonic velocity and compressive strength for self-compacting concrete. Russian. J. Nondestruct. Test.
**2008**, 44, 367–374. [Google Scholar] [CrossRef] - Da Silva, P.R.; de Brito, J. Experimental study of the porosity and microstructure of self-compacting concrete (SCC) with binary and ternary mixes of fly ash and limestone filler. Constr. Build. Mater.
**2015**, 86, 101–112. [Google Scholar] [CrossRef] - Gesoglu, M.; Güneyisi, E.; Özturan, T.; Özbay, E. Modeling the mechanical properties of rubberized concretes by neural network and genetic programming. Mater. Struct. Construct.
**2010**, 43, 31–45. [Google Scholar] [CrossRef] - Liu, M. Self-compacting concrete with different levels of pulverized fuel ash. Constr. Build. Mater.
**2010**, 24, 1245–1252. [Google Scholar] [CrossRef] - Siddique, R.; Aggarwal, P.; Aggarwal, Y. Influence of water/powder ratio on strength properties of self-compacting concrete containing coal fly ash and bottom ash. Constr. Build. Mater.
**2012**, 29, 73–81. [Google Scholar] [CrossRef] - Güneyisi, E.; Gesoglu, M.; Al-Goody, A.; Ipek, S. Fresh and rheological behavior of nanosilica and fly ash blended self-compacting concrete. Constr. Build. Mater.
**2015**, 95, 29–44. [Google Scholar] [CrossRef] - El-Chabib, H.; Syed, A. Properties of self-consolidating concrete made with high volumes of supplementary cementitious mate- rials. J. Mater. Civil. Eng.
**2013**, 25, 1579–1586. [Google Scholar] [CrossRef] - Siad, H.; Mesbah, H.; Mouli, M.; Escadeillas, G.; Khelafi, H. Influence of mineral admixtures on the permeation properties of self-compacting concrete at different ages. Arab. J. Sci. Eng.
**2014**, 39, 3641–3649. [Google Scholar] [CrossRef] - Zhao, H.; Sun, W.; Wu, X.; Gao, B. The properties of the self-compacting concrete with fly ash and ground granulated blast furnace slag mineral admixtures. J. Clean. Prod.
**2015**, 95, 66–74. [Google Scholar] [CrossRef] - Gesoglu, M.; Özbay, E. Effects of mineral admixtures on fresh and hardened properties of self-compacting concretes: Binary, ternary and quaternary systems. Mater. Struct. Construct.
**2007**, 40, 923–937. [Google Scholar] [CrossRef] - Chang, C.C.; Lin, C.J. LIBSVM: A library for support vector machines. ACM Trans. Intell. Syst. Technol.
**2011**, 2, 1–27. [Google Scholar] [CrossRef]

**Figure 3.**Crossover and mutation operation in MGGP: (

**a**) random selection of parent tree nodes; (

**b**) exchange of parents’ genetic material; (

**c**) random node selection in tree mutation; and (

**d**) mutation of a randomly selected part of a tree.

**Figure 4.**Example of the segmentation of variable spaces into regions and 3D regression surfaces for the created regression tree [16].

**Figure 5.**Bootstrap aggregation (bagging) in regression tree ensembles [25].

**Figure 6.**Gradient boosting in regression tree ensembles [25].

**Figure 11.**Comparison of different accuracy criteria for MGGP model as a function of gene number and tree depth (

**a**) RMSE, (

**b**) MAE, (

**c**) MAPE, and (

**d**) R.

**Figure 12.**Display of models that make up the Pareto front marked with green circles, while the optimal model (Model ID 269) is marked with a red circle.

**Figure 18.**Comparison of the accuracy criteria for MLP-ANNs with different numbers of neurons in the hidden layer: (

**a**) RMSE and MAE; and (

**b**) R and MAPE.

**Figure 19.**Comparison of accuracy criteria for the ensemble with a different number of individual ANN models within the ensemble: (

**a**) RMSE and MAE; and (

**b**) R and MAPE.

**Figure 20.**(

**a**) RMSE value for each of the iterations; and (

**b**) the optimal number of neurons in the hidden layer in each of iteration.

**Figure 21.**Comparative analysis of anensemble of ANNs (blue color) and individual ANN models (yellow color) in relation to target test values (red color) for SCC with Class F fly ash compressive strength.

**Table 1.**Review of the application of ML algorithms in the development of compressive strength (CS) prediction models in self-compacting concrete (SCC).

Type of Concrete | Algorithm | Data Points | Year | Authors | Reference |
---|---|---|---|---|---|

SCC | ANN | 80 | 2011 | Siddique et al. | [7] |

SCC | ANN | 169 | 2016 | Asteris et al. | [8] |

SCC | ANN | 114 | 2016 | Douma et al. | [9] |

SCC | ANN | 205 | 2017 | Asteris et al. | [10] |

SCC | Support vector machines (SVM) | 115 | 2019 | Saha et al. | [11] |

SCC | SVM | 340 | 2019 | Azimi-Pour et al. | [12] |

Lightweight self-compacting concrete (LWSCC) | Beetle antennae search (BAS)-algorithm-based random forest (RF) | 131 | 2019 | Zhang et al. | [13] |

SCC | Regression trees (RT), gene-expression programming (GEP), boosting regressor (BR) | 97 | 2021 | Song et al. | [14] |

Self-compacting concrete with waste rubber (SCRC) | Gaussian process regression (GPR) | 144 | 2021 | Hadzima-Nyarko et al. | [15] |

Self-compacting concrete with waste rubber (SCRC) | NN, Bagged trees, RF, boosted trees, SVM, GPR | 166 | 2021 | Kovacevic et al. | [16] |

SCC | NN, NN + firefly optimization algorithm (FOA) | 327 | 2021 | Pazouki et al. | [17] |

SCC | NN, SVM, GEP | 300 | 2021 | Farooq et al. | [18] |

SCC with recycled aggregates | RF, KNN, ERT, XGB, GB, LGBM, CB, GAM | 515 | 2022 | de-Prado-Gil et al. | [19] |

Statistical Analysis of Input and Output Parameters for All Data | ||||||

Constituent | Max. | Min. | Mean | Mode | St.Dev. | Count |

Cement (C) (kg/m^{3}) | 503 | 61.00 | 293.08 | 250.00 | 89.78 | 327 |

Water (W) (kg/m^{3}) | 390.39 | 132.00 | 197.00 | 180.00 | 37.62 | 327 |

Fly ash (A) (kg/m^{3}) | 373.00 | 20.00 | 170.23 | 160.00 | 69.68 | 327 |

Coarse aggregate (CA) (kg/m^{3}) | 1190.00 | 590.00 | 828.34 | 837.00 | 137.30 | 327 |

Fine aggregate (FA) (kg/m^{3}) | 1109.00 | 434.00 | 807.47 | 910.00 | 135.80 | 327 |

Superplasticizer (SP) (%) | 4.60 | 0 | 0.980 | 0.50 | 1.11 | 327 |

Age of samples (AS) (days) | 365 | 1 | 44.31 | 28.00 | 63.76 | 327 |

Compressive strength (MPa) | 90.60 | 4.44 | 36.45 | 12.00 | 19.07 | 327 |

Statistical analysis of input and output parameters for Training set | ||||||

Max. | Min. | Mean | Mode | St.Dev. | Count | |

Cement (C) (kg/m^{3}) | 503.00 | 61.00 | 292.57 | 250.00 | 86.29 | 262 |

Water (W) (kg/m^{3}) | 390.39 | 133.20 | 197.83 | 180.00 | 39.09 | 262 |

Fly ash (A)(kg/m^{3}) | 373.00 | 20.00 | 169.40 | 160.00 | 67.80 | 262 |

Coarse aggregate (CA) (kg/m^{3}) | 1190.00 | 590.00 | 825.08 | 837.00 | 129.38 | 262 |

Fine aggregate (FA) (kg/m^{3}) | 1109.00 | 434.00 | 811.44 | 910.00 | 128.91 | 262 |

Superplasticizer (SP) (%) | 4.60 | 0 | 0.98 | 0.50 | 1.10 | 262 |

Age of samples (AS) (days) | 365 | 1 | 43.86 | 28.00 | 62.63 | 262 |

Compressive strength (MPa) | 90.60 | 4.90 | 36.55 | 12.00 | 19.05 | 262 |

Statistical analysis of input and output parameters for Test set | ||||||

Cement (C) (kg/m^{3}) | 503.00 | 61.00 | 295.11 | 295.11 | 103.36 | 65 |

Water (W) (kg/m^{3}) | 279.50 | 132.00 | 193.68 | 193.68 | 31.07 | 65 |

Fly ash (A)(kg/m^{3}) | 336.00 | 20.00 | 173.56 | 173.56 | 77.27 | 65 |

Coarse aggregate (CA) (kg/m^{3}) | 1190.00 | 590.00 | 841.48 | 841.48 | 165.95 | 65 |

Fine aggregate (FA) (kg/m^{3}) | 1109.00 | 434.00 | 791.43 | 791.43 | 160.74 | 65 |

Superplasticizer (SP) (%) | 4.60 | 0 | 0.98 | 0.98 | 1.15 | 65 |

Age of samples (AS) (days) | 365 | 1 | 46.09 | 46.09 | 68.61 | 65 |

Compressive strength (MPa) | 72.61 | 4.44 | 36.02 | 4.44 | 19.32 | 65 |

Parameter | Setting |
---|---|

Function set | times, minus, plus, rdivide, square, exp, log, mult3, sqrt, cube, power |

Population size | 100 |

Number of generations | 1000 |

Max number of genes | 6 |

Max tree depth | 7 |

Tournament size | 10 |

Elitism | 0.05% of population |

Crossover probability | 0.85 |

Mutation probability | 0.1 |

Probability of Pareto tournament | 0.5 |

Model ID | RMSE | MAE | MAPE | R |
---|---|---|---|---|

901 | 8.3826 | 6.3119 | 0.2437 | 0.9065 |

316 | 8.6407 | 6.5785 | 0.2578 | 0.9038 |

320 | 7.9860 | 6.2587 | 0.2444 | 0.9175 |

236 | 7.2654 | 5.7730 | 0.2411 | 0.9294 |

269 | 7.0962 | 5.6960 | 0.2409 | 0.9295 |

Min Leaf Size | RMSE | MAE | MAPE | R |
---|---|---|---|---|

1 | 9.5144 | 7.2558 | 0.2687 | 0.8873 |

2 | 9.5368 | 7.3382 | 0.2821 | 0.8780 |

3 | 9.0359 | 6.9911 | 0.2627 | 0.8873 |

4 | 9.4848 | 7.3640 | 0.2753 | 0.8775 |

5 | 9.6157 | 7.3791 | 0.2769 | 0.8747 |

6 | 8.8407 | 6.6424 | 0.2253 | 0.8942 |

7 | 9.7815 | 6.9879 | 0.2458 | 0.8685 |

8 | 10.1599 | 7.3942 | 0.2640 | 0.8569 |

9 | 10.9231 | 8.4503 | 0.3106 | 0.8283 |

10 | 10.7970 | 8.4804 | 0.3232 | 0.8312 |

SVM linear | $C=0.1428$ | $\epsilon =0.0436$ | / |

SVM RBF | $C=17.6919$ | $\epsilon =0.0412$ | $\gamma =2.1220$ |

SVM sigmoid | $C=252.3998$ | $\epsilon =0.0547$ | $\gamma =9.1574$ |

GP Model Covariance Function | Covariance Function Parameters | ||
---|---|---|---|

Exponential | $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}exp\left[-\frac{1}{2}\frac{r}{{\sigma}_{l}{}^{2}}\right]$ | ||

${\sigma}_{l}=$ 31.24 | ${\sigma}_{f}=$ 54.56 | ||

Squared exponential | $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}exp\left[-\frac{1}{2}\frac{{\left({x}_{i}-{x}_{j}\right)}^{\mathrm{T}}\left({x}_{i}-{x}_{j}\right)}{{\sigma}_{l}{}^{2}}\right]$ | ||

${\sigma}_{l}=$ 1.71 | ${\sigma}_{f}=$ 29.25 | ||

Matérn 3/2 | $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}\left(1+\frac{\sqrt{3}r}{{\sigma}_{l}}\right)exp\left[-\frac{\sqrt{3}r}{{\sigma}_{l}}\right]$ | ||

${\sigma}_{l}=$ 4.17 | ${\sigma}_{f}=$ 41.74 | ||

Matérn 5/2 | $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}\left(1+\frac{\sqrt{5}r}{{\sigma}_{l}}+\frac{5{r}^{2}}{3{\sigma}_{l}{}^{2}}\right)exp\left[-\frac{\sqrt{5}r}{{\sigma}_{l}}\right]$ | ||

${\sigma}_{l}=$ 2.68 | ${\sigma}_{f}=$ 34.12 | ||

Rational quadratic | $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}{\left(1+\frac{{r}^{2}}{2a{\sigma}_{l}{}^{2}}\right)}^{-\alpha};r=0$ | ||

${\sigma}_{l}=$ 3.05 | $a=$ 0.22 | ${\sigma}_{f}=$ 47.62 |

Covariance Function Parameters | ||||||
---|---|---|---|---|---|---|

${\mathit{\sigma}}_{1}$ | ${\mathit{\sigma}}_{2}$ | ${\mathit{\sigma}}_{3}$ | ${\mathit{\sigma}}_{4}$ | ${\mathit{\sigma}}_{5}$ | ${\mathit{\sigma}}_{6}$ | ${\mathit{\sigma}}_{7}$ |

ARD exponential: $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}\mathrm{exp}\left(-r\right)$$;{\sigma}_{F}$$=64.78;r=\sqrt{{\displaystyle \sum}_{m=1}^{d}\frac{{\left({x}_{im}-{x}_{jm}\right)}^{2}}{{\sigma}_{m}{}^{2}}}$ | ||||||

109.27 | 61.97 | 97,176.47 | 74.09 | 24.72 | 31.09 | 10.95 |

ARD squared exponential: $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}exp\left[-\frac{1}{2}{\displaystyle \sum}_{m=1}^{d}\frac{{\left({x}_{im}-{x}_{jm}\right)}^{2}}{{\sigma}_{m}{}^{2}}\right];$${\sigma}_{f}$ = 32.48 | ||||||

4.12 | 3.35 | 7.92 | 6.31 | 3.29 | 0.48 | 1.12 |

ARD Matérn 3/2: $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}\left(1+\sqrt{3}r\right)exp\left[-\sqrt{3}r\right];$${\sigma}_{f}$ = 25.12 | ||||||

4.61 | 2.33 | 5649.90 | 3.39 | 0.97 | 1.37 | 0.48 |

ARD Matérn 5/2: $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}\left(1+\sqrt{5}r+\frac{5{r}^{2}}{3}\right)exp\left[-\sqrt{5}r\right];$${\sigma}_{f}$ = 30.34 | ||||||

4.23 | 4.86 | 8.88 | 6.91 | 3.99 | 0.70 | 0.53 |

ARD rational quadratic: $k\left(\left({x}_{i},{x}_{j}|\mathsf{\Theta}\right)\right)={\sigma}_{f}^{2}{\left(1+\frac{1}{2\alpha}{\displaystyle \sum}_{m=1}^{d}\frac{{\left({x}_{im}-{x}_{jm}\right)}^{2}}{{\sigma}_{m}{}^{2}}\right)}^{-\alpha};\alpha $$=30.34;{\sigma}_{f}$ = 51.96 | ||||||

4.45 | 1.97 | 18,417.04 | 4.87 | 1.08 | 5.89 | 0.57 |

Parameter | Parameter Value | |
---|---|---|

Lower Limit | Upper Limit | |

Number of epochs | / | 1000 |

MSE value (performance) | / | 0 |

Gradient | / | 1.00 × 10^{−7} |

$\mathrm{The}\mathrm{value}\mathrm{of}\mathrm{the}\mathrm{parameter}{\lambda}_{k}$ (Mu) | 0.005 | 1.00 × 10^{10} |

Model | RMSE | MAE | MAPE/100 | R |
---|---|---|---|---|

MGGP | 7.0962 | 5.6960 | 0.2409 | 0.9295 |

Decision tree | 8.8407 | 6.6424 | 0.2253 | 0.8942 |

TreeBagger | 7.0236 | 5.5892 | 0.2387 | 0.9378 |

Random forest | 6.9324 | 5.5627 | 0.2379 | 0.9389 |

Boosted tree 1 | 5.9597 | 4.8307 | 0.1793 | 0.9518 |

Boosted tree 2 | 6.3814 | 4.5410 | 0.1580 | 0.9237 |

SVM linear | 12.7268 | 10.6332 | 0.5323 | 0.7495 |

SVM RBF | 5.9533 | 4.5551 | 0.1976 | 0.9521 |

SVM sigmoid | 12.6875 | 10.4926 | 0.5242 | 0.7511 |

GP exponential | 6.7391 | 5.3043 | 0.2313 | 0.9376 |

GP Sq.exponential | 6.5298 | 5.0244 | 0.2117 | 0.9429 |

GP Matérn 3/2 | 6.3409 | 4.7695 | 0.1909 | 0.9454 |

GP Matérn 5/2 | 6.3686 | 4.7852 | 0.1943 | 0.9452 |

GP Rat. quadratic | 6.4138 | 4.8693 | 0.1970 | 0.9439 |

GP ARD exponential | 5.9891 | 4.4334 | 0.1625 | 0.9517 |

GP ARD Sq. exponential | 6.2278 | 4.6602 | 0.1739 | 0.9506 |

GP ARD Matérn 3/2 | 6.2476 | 4.6911 | 0.1559 | 0.9481 |

GP ARD Matérn 5/2 | 6.4760 | 4.5837 | 0.1632 | 0.9463 |

GP ARD Rat. quadratic | 6.2025 | 4.6228 | 0.1560 | 0.9494 |

ANN | 8.8034 | 6.6700 | 0.2783 | 0.8978 |

Ensemble ANN | 5.6351 | 4.3665 | 0.1734 | 0.9563 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Kovačević, M.; Lozančić, S.; Nyarko, E.K.; Hadzima-Nyarko, M.
Application of Artificial Intelligence Methods for Predicting the Compressive Strength of Self-Compacting Concrete with Class F Fly Ash. *Materials* **2022**, *15*, 4191.
https://doi.org/10.3390/ma15124191

**AMA Style**

Kovačević M, Lozančić S, Nyarko EK, Hadzima-Nyarko M.
Application of Artificial Intelligence Methods for Predicting the Compressive Strength of Self-Compacting Concrete with Class F Fly Ash. *Materials*. 2022; 15(12):4191.
https://doi.org/10.3390/ma15124191

**Chicago/Turabian Style**

Kovačević, Miljan, Silva Lozančić, Emmanuel Karlo Nyarko, and Marijana Hadzima-Nyarko.
2022. "Application of Artificial Intelligence Methods for Predicting the Compressive Strength of Self-Compacting Concrete with Class F Fly Ash" *Materials* 15, no. 12: 4191.
https://doi.org/10.3390/ma15124191