Predicting the compressive strength of concrete with fly ash admixture using machine learning algorithms

• Published in: Construction & building materials 2021-11, Vol.308, p.125021, Article 125021
• Main Authors: Song, Hongwei, Ahmad, Ayaz, Farooq, Furqan, Ostrowski, Krzysztof Adam, Maślak, Mariusz, Czarnecki, Slawomir, Aslam, Fahid
• Format: Article
• Language: English
• Subjects: Artificial neural network; Bagging regressor; Compressive strength; Concrete; Decision tree; Fly ash; Gene expression programming; Individual algorithms; Predictions
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2021.125021

[Display omitted] •Comparative analyses of machine learning tools for compressive strength prediction.•Self-compacting concrete with high amount of fly ash.•Eco-friendly technique of investigating SCC compressive strength. The cementitious composites have different properties in the changing environment. Thus, knowing their mechanical properties is very important for safety reasons. The most important in the case of concrete is the Compressive strength (CS). To predict the CS of concrete Machine learning (ML) approaches has been essential. This study includes the collection of data from the experimental work and the application of ML techniques to predict the CS of concrete containing fly ash. The chemical and physical properties of all the materials used in this study were evaluated. Although, the emphasis of this research is on the use of supervised machine learning algorithms to forecast the CS of concrete. The Gene expression programming (GEP), Artificial neural network (ANN), and Decision tree (DT) algorithms were investigated for the prediction of outcome (CS). Concrete samples (cylinders) with different mix ratios were cast and tested at various ages to maintain the required data for applying it to run the models. Total 98 data points were collected from the experimental approach, in which seven parameters namely cement, fly ash, superplasticizer, coarse aggregate, fine aggregate, water, and days were taken as input to predict the output which was CS parameter. The experimental data is further validated by mean of k-fold cross-validation using R2, root mean error (RME), and Root mean square error (RMSE). In addition, statistical checks were incorporated to evaluate the model performance. In comparison, the bagging algorithm shows high accuracy towards the prediction of outcome as indicated by its high coefficient correlation (R2) value equals to 0.95, while R2 value for GEP, ANN and DT comes to 0.86, 0.81 and 0.75 respectively.