Fig. 4
From: OCT-based diagnosis of glaucoma and glaucoma stages using explainable machine learning
(a) Partial dependency plot (PDP) comparison of early, moderate, and advanced glaucoma for RNFL Symmetry (b) PDP for RNFL Inferior (c) PDP and individual conditional expectation (ICE) Plots for RNFL symmetry (estimated cut-off = 71%), (d) PDP and ICE plots for RNFL inferior (estimated cut-off = 88.13 μm), (e) two-way numerical PDP for RNFL symmetry and RNFL inferior (f) 3D feature interaction plot for RNFL symmetry and RNFL inferior. In all cases (a–f), the y-axis represents the predicted probability of a machine learning model, and the x-axis presents the magnitude of feature values. For Individual Conditional Expectation (ICE) plots (b,d), the thin separate curves show the dependency of the prediction on the feature (individual dependence from each sample) and the thick curves represent the average effect (mean partial dependence).
