Fig. 6

Error variation under progressive feature corruption. Note: \({\sigma }\) denotes feature perturbation intensity (0 → 1 corresponds to random masking ratio of high-importance features increasing from 0 to 100%). Metrics include MAE (Mean Absolute Error), \({\text{R}}^{2}\) (Coefficient of Determination), and ACC (Accuracy), with values reflecting model tolerance to feature corruption.