Comparison of traditional and automated machine learning approaches in predicting the compressive strength of graphene oxide/cement composites

Abstract

The prediction of the compressive strength (CS) of graphene oxide reinforced cement composites (GORCCs) is crucial for accelerating their potential application in civil engineering. However, traditional experimental and theoretical modelling suffer from problems such as time-consuming, costly, and inefficient, etc. It is also challenging to consider the effects of multiple coupling factors. In this work, machine learning (ML) approaches are developed as the first attempt to explore the complex relationships between the CS of GORCCs and the multiple coupling factors. A comprehensive dataset of 260 experimental results is collected to train and test the ML models. It is demonstrated that the developed model can accurately predict the CS of GORCCs. The feature importance analysis reveals that dispersion of sonicating GO in polycarboxylate superplasticizer solution is the most favorable dispersion method for achieving good dispersion. Among the ML models used, it is found that the AutoGluon-Tabular (AGT) model not only demonstrates the highest confidence in predictions but also offers better interpretability of the results. Moreover, users can train AGT models more efficiently compared to traditional ML workflows, avoiding the time-consuming process of hyperparameter tuning.