Prediction of carbon nanotube atomic coordinates based on machine learning algorithms

In this thesis, seven prediction models (i.e. Feed Forward Neural Network (FFNN), Function Fitting Neural Network (FITNET), Cascade-Forward Neural Network (CFNN), Generalized Regression Neural Network (GRNN), Support Vector Regression (SVR), Classification and Regression Tree (CART) and Multiple Linear Regression (MLR)) have been developed for atomic coordinate prediction of carbon nanotubes. The main aim of this research is to reduce the calculation time for atomic coordinates from days to minutes using developed prediction models. The dataset was created by combining the atomic coordinates of elements and chiral vectors using CASTEP software. The accuracy of the proposed models is evaluated with Mean Square Error (MSE), Mean Absolute Error (MAE), Standard Error of the Estimate and Correlation Coefficient metrics. The dataset is studied separately with and without using 10-fold cross-validation. FFNN, CFNN and FITNET prediction models yielded very high performance by means of MSE and MAE. These models are followed by CART, SVR, GRNN and MLR, respectively. The results obtained from this study can be used in two ways: i) The atomic coordinates can be predicted with a low-error without using a simulation program, ii) The estimated results can be used as an initial value of simulation software for reducing duration of the atomic coordinate simulation seriously.