Fig. 7: Feature transferability.

Box plots illustrate RMSE for a CaO, b BaO, and c ZnO applying the MgO-derived features. We note that RMSE is obtained from models with five descriptors even if the method selects more than five descriptors to have a fair comparison between the SL methods because Horseshoe prior usually selects less than five features (Fig. 10). Comparison of models for describing d CaO, e BaO, and f ZnO data built from features selected by LASSO, Horseshoe prior, and Dirichlet–Laplace prior using the MgO training data in Fig. 5. RMSE is computed using the l₀ norm method to refine the total descriptor set identified with each method to a model containing the number of descriptors shown on the x-axis. The center line, upper bound, lower bound, upper whisker, and lower whisker in box plots represent median, 75th percentile, 25th percentile, maximum, and minimum, respectively.