Machine learning approaches to predict the strength of graphene nanoplatelets concrete: Optimization and hyper tuning with graphical user interface

This paper focuses on the need to consider the use of multipurpose cement composite (CC) in the construction industry to give superior performance. Therefore, incorporating nanomaterials (NMs) can improve the performance and characteristics of CC. Hence, the use of graphene nanoplatelets (GNPs) in the composite matrix can be a viable way to address the challenges towards achieving a sustainable, with superior properties. In addition, forecasting the properties of NMs is a challenging task due to nature of ambiguity between parameters, complex nature, and non-linear response to strength. In addition, strength evaluation is time consuming process. Thus, there is a needs for predictive models to estimate strength of NMs. This study employee four machine learning (ML) approaches namely as light gradient boosting (LGB), artificial neural network (ANN), gradient boosting (GB), and k-nearest neighbor (KNN) with hyper-tuning, and optimization. In addition, model evaluation is judge by statistical measures, uncertainty analysis, and tenfold approach is applied for validation of the model. Moreover, graphical user interface (GUI) is developed for practical implementation to estimate strength. The result reveals that XGB, and ANN model shows robust analysis with greater R-2 > 0.90 for both train and test sets with XGB perform better as compared to ANN and other models. XGB depicts lesser statistical index, and uncertainty analysis demonstrates all model with less level of ambiguity for train and test set. The train and test set for XGB, LGB, KNN, and ANN models demonstrate 7.668%, 8.9%, 8.9%, 17.18%, 9.85%, 20.61, and 5%, 14.74%, respectively. Shapley analysis reveals that GNP thickness, diameter and curing have major contribution to strength.