Efficient prediction of mechanical characteristics of single wall carbon nanotubes

Single wall carbon nanotubes (SWCNTs) have garnered significant attention due to their exceptional mechanical properties, which vary with chiral indices (n, m). While molecular dynamics (MD) simulations provide accurate assessments, they are computationally expensive. Machine learning (ML) methods offer a faster alternative while maintaining high predictive accuracy. In this study, we employed a random forest (RF) model on two datasets of pristine SWCNTs, utilizing the chiral index (n) and chiral angle $(\theta)$ as features. The RF model applied to the first dataset achieved test root mean squared error (RMSE) values of 0.0028, 0.0056, and 0.0029 for predicting Young’s modulus, fracture strain, and ultimate strength, respectively. In contrast, the RF model on the second dataset yielded RMSE values of 0.0027,0.0055, and 0.003 for these parameters. By fitting the predicted values of the second dataset into a second-order polynomial, we obtained stressstrain curves with an average $R^{2}$ of 0.9999 and an RMSE of 0.006, surpassing prior neural network models while significantly reducing computational time.