Estimation of the compressive strength of green concretes containing rice husk ash: a comparison of different machine learning approaches

Abstract

To mitigate the environmental issues related to the utilisation of ordinary portland cement (OPC) in concrete mixtures, attempts have been carried out to find alternative binders such as rice husk ash (RHA) as replacements for OPC. This study contributes to moving from the traditional laboratory-based methods for the determination of compressive strength (CS) towards machine learning-based approaches by developing three accurate models (i.e. artificial neural network (ANN), multivariate adaptive regression spline (MARS) and M5P model tree) for the estimation of the CS of concretes containing RHA. For this purpose, the models were developed employing 909 data records collected through technical literature. The results indicate that all three techniques provide reliable estimations of the CS for both training and testing datasets. However, the ANN-based model outperforms the other two models, while the MP5-based model is associated with the least accuracy and the maximum error values among all three techniques. The parametric study revealed that by increasing the contents of cement, coarse aggregate and age, and decreasing the contents of water, fine aggregate, RHA and superplasticizer, the CS of concrete increases while the sensitivity analysis demonstrated that the coarse aggregate content was the most influential parameter affecting the values of CS.

Keywords:

• Machine learning
• concrete
• rice husk ash
• compressive strength
• parametric study
• sensitivity analysis

Author contributions

Amir Tavana Amlashi: Conceptualisation, Data curation, Methodology and Writing–Original Draft. Emadaldin Mohammadi Golafshani: Methodology and Writing–review and editing. Seyed Abolfazl Ebrahimi: Data curation and Methodology. Ali Behnood: Methodology, Writing–Original Draft, Writing–review and editing and supervising.

Availability of data and material

The authors confirm that the data supporting the findings of this study are available within the documents discussed in ‘Data collection’ section.

Disclosure statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Funding

No funding was received for this project.