Prediction of lattice constant of pyrochlore compounds using optimized machine learning model

The pyrochlore compounds has a general formula of A2 under cubic structure and have been identified as a good catalyst for production of clean energy due to its unique physical properties. In this study, we used an optimized machine learning technique based on Particle Swarm Optimization-Support Vector Regression (PSO-SVR) to learn the relationship between its structural properties with its lattice constants. We consider the ionic radii and electronegativities values of each cation and anion constituted in the crystal structure as the input variables. The dataset includes 220 data of different simple and mixed ternary pyrochlore compounds that are ranging from 9.2870 until 10.6820 Å which is larger than previous researchers. Our study revealed that, the selected input variables are effectively predicted the lattice constants of pyrochlore compounds. The results shows that as we increase the number of input factors in the prediction, the performance of the prediction is increases. Three different kernel functions were used in PSO-SVR (Linear, Polynomial, and RBF kernel) shows that PSO-SVR algorithm with RBF function had better accuracy than other kernel functions. Furthermore, we compare the accuracy of our PSO-SVR algorithm with other previous studies method such as BSVR, ANN, and linear regression. The analysis revealed that, our model displayed a better performance compared to previous researchers with RMSE of 0.0122 Å. Hence, the developed PSO-SVR model allows a new access for precisely validate the pyrochlore compounds crystal structure properties.

Actions (login required)

View Item