Density of Bismuth Boro Zinc Glasses Using Machine Learning Techniques

Machine learning techniques have been employed to predict the glass densities of xBi(2)O(3)–(70 − x)B(2)O(3)–20Li(2)O–5Sb(2)O(3)–5ZnO glasses using a data set of 2000 various B(2)O(3) rich glasses using their chemical composition and ionic radius. The experimental density of present glasses strongly depends on Bi(2)O(3) content which is increasing with bismuth content. The increasing density in bismuth doped glasses because the BO(3) are converted into BO(4) units, and besides BO(3) units are less heavy than the BO(4) units. The FTIR studies also confirm that the intensity of B–O–B bond decreasing with increasing Bi(2)O(3) content which suggested that B–O–B bond in bond ring isolated to BO(3) units transformed into BO(4) units. In Raman Spectra the stretching vibrations of BO(4) units shifting towards higher wavelengths with increasing Bi(2)O(3) content. This shifting conforms that there is a structural changes in the glass-matrix and borate units converting from BO(3) to BO(4) units. The prepared glasses along with B(2)O(3) rich glass data set train on various AI model such as gradient descent, Random Forest regression and Neural Networks to predict present density of glasses. Among the various models RF regression analysis model is successfully acceptable for the glass data with the highest R(2) value 0.983 which end result conform that the predicted and experimental values correlated. ANNs stood the effective technique in prediction of glass density with the optimum performance resulting with Tanh as the activation function (R(2) = 0.950). The minimum cost 0.018 obtained in the case of gradient decent function which also shows the better performance of regression model.