Fabrication, physical and machine learning density prediction techniques of newly B2O3–ZnO–BaO–PtO2 glasses

Abstract

New glass samples with chemical formula (50-x)B2O3–25ZnO–25BaO-xPtO2 (0 (Pt-0.0) ≤ x ≤ 1 (Pt-1.0) mol%) have been fabricated for first time using the traditional technique of melt-quenching. The significant effect of PtO2 on the physical properties of the Pt-glasses has been investigated. Various machine learning algorithms (MLAs) have been applied for corresponding density prediction using their chemical composition. The density (ρglass) of Pt-glasses was enhanced from 3.05 g/cm3 to 4.10 g/cm3. Using K-Nearest Neighbors (KNN) algorithm, the predicted values of densities were close to the experimental values, with an accuracy of 0.825 and an R2 value of 0.800. In the artificial neural network (ANN) algorithm, applying the activation function of ‘tanh’, predicted the glass density closest to the experimental values with an R2 value of 0.890. In the support vector machines (SVM) algorithm, the results showed that the polynomial kernel with a degree of 3 produced the best fit for the data. The R2 value for SVM is 0.920. The random forest regression (RFR) has proven to be more dominant while predicting density of present glasses followed by SVM with higher R2 value 0.945 which shows the best fit for glass data.