Extreme learning machine and swarm-based support vector regression methods for predicting crystal lattice parameters of pseudo-cubic/cubic perovskites

Abstract

Lattice parameters of perovskite compounds play crucial roles in engineering of buffer layers and substrates for heteroepitaxial films. As a result, predictive models that can effectively estimate these lattice parameters are highly desired. Therefore, this work presents elegant methods of predicting lattice parameters of pseudo-cubic/cubic perovskite through the development of the extreme learning machine (ELM) based model and hybridization of the particle swarm optimization (PSO) technique with the support vector regression algorithm (SVRA). The generalization and predictive strengths of the proposed SVRA-PSO and ELM-based models are compared with existing methods such as the Ubic model and the recently developed Sidey model on the basis of root mean square error (RMSE), mean absolute error, mean absolute percentage error (MAPE), and correlation coefficient. The developed SVRA-PSO model performs better than the ELM-based model, the Ubic model, and the Sidey model, with performance improvement of 20.99%, 29.29%, and 33.39%, respectively, on the basis of MAPE. Similarly, the SVRA-PSO, respectively, attains performance improvement of 24.74%, 34%, and 37.89% on the basis of RMSE. Furthermore, the developed ELM-based model outperforms the Ubic and Sidey models with performance improvement of 15.70% and 10.50%, respectively, on the basis of MAPE and percentage enhancement of 17.48% and 12.31% when compared on the basis of RMSE. Although the SVRA-PSO model has the best performance of all the compared models, the developed ELM-based model possesses the advantage of easy implementation in addition to its moderate performance.