Towards accelerating the development of calcined clay cements: data-driven prediction of compressive strength exploiting machine learning algorithms

Calcined clay cements can potentially reduce the carbon footprint of cement production sector. However, the prediction of engineering properties of this low carbon cement is still inaccurate using empirical methods due to the considerable number of parameters involved during formulation and hardening. By the exploitation of the machine learning (ML) approach, this paper aims to predict the influence of material composition and hardening conditions on the compressive strength of calcined clay cement. For this purpose, seven algorithms of supervised ML were deployed. XGboost algorithm was identified as the best one to predict the compressive strength with a mean absolute error of 2.5 MPa. Results show that alumina to silica ratio is the most impacting input feature on the development of the cement strength. Furthermore, the study highlights interesting potential formulations for further optimizations