Fig. 1. Genetic algorithm workflow^[23]

Fig. 2. Artificial neural network mode

Fig. 3. Bayesian interpolation process using ML^[36]

Fig. 4. Comparison between machine learning and conjugate gradient optimization^[36]. (a) Contour plot showing the cost function R_χ as a function of q_si and A_SiO, the red and black circles represent the paths explored by machine learning and conjugate gradient optimization, respectively; (b) scaling of the data in Fig. (a); (c) evolution of cost function R_χ in machine learning and conjugate gradient optimization (inset: enlarged data obtained under conjugate gradient optimization)

Fig. 5. Boosting Component Property Correlation Prediction^[37]. (a) Pure data-driven prediction using linear regression, random forest, and artificial neural network methods; (b) "physics-based" prediction

Fig. 6. Scatter plot of predicted leaching rate^[37]. (a) Linear regression; (b)Lasso regression; (c) elastic network regression; (d) support vector machine regression; (e) random forest; (f) artificial neural network algorithm

Fig. 7. Forecast results^[39]. (a) Average influence of SHAP average on model output; (b) summary of V_d SHAP values

Fig. 8. Predicted values and their histograms values of (a) refractive index and (b) glass transition temperature (inset: histogram of forecast residuals, the difference between reported and predicted values, the vertical color bars show the density of the data points)^[45]

Fig. 9. Mean and standard deviation of predicted residuals for each chemical element in the data set^[45]. (a) Refractive index; (b) glass transition temperature (numbers in parentheses are the number of glass components containing the chemical element in the retained data set, and the predicted residual is the difference between the reported and predicted values, with the elements in order from left to right and from less to more)