A novel approach in forecasting compressive strength of concrete with carbon nanotubes as nanomaterials

The evolution of nanotechnology in cementitious concrete can be used to enhance the mechanical behavior of concrete and other materials. Thus, the utilization of carbon nanotubes (CNTs) in the concrete cementitious industry will deliver important remedial measures for generating an eco-friendly environment with multifunctional properties. In addition, planning and conducting experimental tests for different samples and testing at varying ages is laborious, time taking and expensive. Consequently, the prediction model looks necessary. Because of the complex nature of contemporary construction materials, empirical and statistical approaches do not work well for such materials; thus, machine learning approaches were utilized for this study. Thus, the process of forecasting concrete strength (CS) can be accelerated by employing soft machine learning (ML) techniques. Moreover, the ensemble of the models is also done for an accurate estimation of the CS with the Python Jupyter Notebook. Decision tree (DT), multilayer perceptron neuron network (MLPNN), and support vector machine (SVM) were utilized as individual approaches. These individual ML models then ensemble with two approaches namely bagging and boosting with twenty sub models. The outcome of the model depicts a robust performance by showing significant correlations (R2) as compared to individual ML. Data points with 282 having contribution factors (e.g., curing (days), cement (kg/m3), water to binder (w/b), fine aggregate (kg/m3), nanomaterial as carbon nanotubes (%), coarse aggregate (kg/m3), and CS were used as input parameters for ML modeling. To ensure that each model is as accurate as possible, cross-validation (CV) with K-folding and statistical error analysis (i.e., MAE, MAE, and RMSE) was utilized. The result reveals that the ensemble algorithm with bagging and boosting give robust correlation R2 ranging from 0.9 to 0.95. Moreover, DT with Ada-boost yields a strong R2 of about 0.93. Similarly, ensemble model of SVM and MLPNN with boosting depicts R2 = 0.93 and 0.94 respectively. K-fold cross-validation confirms the model’s accuracy with minimal statistical error. Also, statistical checks reveal that the ensemble model provides an increase in the overall performance by showing minimal error as compared to standalone method.