Data-Driven Prediction of Electrical Resistivity of Graphene Oxide/Cement Composites Considering the Effects of Specimen Size and Measurement Method

Abstract

The prediction of electrical resistivity of graphene oxide (GO) reinforced cement composites (GORCCs) is essential to promote the application of the composites in civil engineering. Traditional experiments find it challenging to capture the effect of various features on the electrical resistivity of the GORCCs. In this work, machine learning (ML) techniques are employed to explore the complex nonlinear relationships between different influencing factors and the electrical resistivity of the GORCCs. A total of 171 datasets are utilized for training and testing the ML models. It is demonstrated that the applied ML models are effective and efficient. Apart from the water/cement ratio, correlation analysis shows that the electrical resistivity of the GORCCs is highly dependent on the specimen size and measurement method. Feature importance analysis shows that the dispersion of GO has a significant influence on the electrical resistivity. The extreme gradient boosting (XGB) model and the artificial neural network (ANN) model with 3 hidden layers are proven to have better predictions, as evidenced by higher R2 and lower root mean square error (RMSE). This work is envisioned to provide an effective and efficient way to identify the complex relationship between the material properties of the GORCCs and the various influencing factors.