fig2

Figure 2. Model performance comparisons across different optimization stages. (A) Heatmap showing prediction accuracy for various models under multiple descriptor filtering methods; (B) Performance metrics for the PI-H dataset after descriptor selection using the F-test and initial HPO with GS; (C) Performance metrics for the PI-F dataset following descriptor selection with VS and GS-based HPO; (D) Enhanced model performance for the PI-H dataset after extensive HPO using Optuna; (E) Improved model performance for the PI-F dataset with Optuna optimization. HPO: Hyperparameter optimization; GS: grid search; VS: variance screening.