Modelling the Optical Band Gap Energy of Undoped ZnO Thin Films by Supervised Machine Learning Methods

Abstract

Recently, machine learning (ML) methods are growing as one of the most powerful techniques in scientific research and technological applications. Herein, artificial neural networks (ANNs) as novel predictive ML techniques were built to predict the optical band gap energy of undoped ZnO thin films. The proposed multilayer perceptron neural network methods include the scaled conjugate gradient (SCG) and the gradient descent (GD) with momentum and learning rate optimization coefficients.

The two suggested techniques were trained, tested, and validated with empirical data sets, by selecting the temperature of the substrate and the precursor molarity of ZnO solution as input parameters as well as the band gap energy as a response parameter. Furthermore, the simulated findings of ANN models were compared to the multiple linear regression (MLR) model and then the fitness and accuracy of those models were evaluated by different statistical metrics including the root mean square error (RMSE), the mean absolute percentage error (MAPE), and regression coefficients. Based on the results, SCG-ANN and GD-ANN models show high prediction accuracy with RMSE of 0.055 and 0.064 for testing data, respectively, whilst MLR analysis showed poor prediction accuracy with RMSE of 0.080 and R² of 0.063. Additionally, the simulated output of these proposed ANNs models is in good agreement with the empirical datasets, which indicates high performance of SCG-ANN and GD-ANN models than the MLR model