Prediction of energy band gap of zno using machine learning approach

Mohd Shazali, Nur Athirah (2022) Prediction of energy band gap of zno using machine learning approach. [Project Paper] (Submitted)

Text FS 2022 58.pdf Download (3MB)

Abstract

Zinc oxide, ZnO is an inorganic compound that has a great property which is wide direct band gap energy. This significant characteristic has contributed to the extensive use of the energy band gap (Eg) of ZnO in many branches of the application area such as solar cells and laser diodes. Traditionally, the measurement of the Eg has been done through experimental procedures that are more complicated and high cost. Therefore, a machine learning approach was used in determination of the Eg of ZnO since the process of synthesizing ZnO properties is time-consuming. Moreover, there is no initial assessment about the correlation of different inputs dataset that influence the prediction of energy band gap of ZnO. Thus, the main aim of this thesis is to develop machine learning models that can accurately predict the energy band gap of ZnO. This study used four different algorithms which are ARIMAX and PSO-SVR models which involve three types of kernel functions that are linear, polynomial, and RBF. The comparison between these methods showed that the development model of the RBF PSO-SVR resulted in lower RMSE and MAE, which 0.0395eV and 0.032eV, respectively. Next, the cross-validation techniques which is k-fold where k is 3, 5, and 10 in the PSO-SVR model were evaluated with the analysis of RMSE and MAE as well as CC. The developments of 5-fold and 10-fold validations suit well with the RBF PSO-SVR model which the RMSE and MAE gained are 0.0863eV and 0.0644eV, respectively. Whereas, the highest CC obtained for 5-fold and 10-fold validations were 0.7943 and 0.7836, respectively. The RBF kernel also performs efficiently in different sizes of split datasets in which 80% of train dataset and one testing data with the RMSE are 0.0434eV and 1.08 x 10-6eV, respectively. The studies of different influence factors represent that lattice constant a along with 136 datasets given the highest value of correlation with energy band gap which is 0.5905. Generally, the development model of PSO-SVR (RBF) performs the best fitting model that is suitable for the 5-fold and 10-fold cross validation techniques. This analysis also showed that the combination of lattice constant a and crystallite size datasets had lower error since lattice constant a has high correlation with Eg.

Item Type:	Project Paper
Faculty:	Fakulti Sains
Depositing User:	Ms. ROHANA ALIAS
Date Deposited:	05 Jun 2024 23:36
Last Modified:	05 Aug 2024 09:12
URI:	http://psaspb.upm.edu.my/id/eprint/1898

Actions (login required)

View Item