Fig. 1: Top 30 features ranked by relative importance during the test phase, as measured by mean SHAP values and frequency of selection. | npj Parkinson's Disease

Fig. 1: Top 30 features ranked by relative importance during the test phase, as measured by mean SHAP values and frequency of selection.

From: Spontaneous eye blink-based machine learning for tracking clinical fluctuations in Parkinson’s disease

Fig. 1

For each target task, 20 blink-related features were selected during training, and their mean SHAP values (bars) and standard deviations are displayed. Bar colours represent the frequency with which each feature was selected across training iterations. Panels display feature importance for A Dyskinesia classification, B ON/OFF classification, C MDS-UPDRS Part III regression, and D Plasma L-dopa concentration regression. Refer to Fig. 4C and Tables 2 and 3 for details on feature categories and extraction processes.

Back to article page