FIGURE 3.

Tree SHAP Plot of Global Feature Importance indicating the composition and cumulative ratios for the 38 baseline predictors tested in the RF classifier model (AUC = 0.85, Accuracy = 0.81, Precision = 0.81, Recall = 0.51, nMCC = 0.76, F₁ = 0.60). Predictors are plotted as their individual composition ratio (blue bars; scale shown at the top of the plot) in descending order of importance. Composition ratio is the amount the predictor contributes to the model output. The blue curved line reflects the cumulative ratio with each added predictor (scale shown at the bottom of the plot). The line arcs from approximately 12% for the most important predictor [gait (steps)] to 100% for the least important predictor (APOE). For example, creatinine (the final predictor indicated in the black rectangle) has a composition ratio of approximately 4 (explains 4% of the model; as indicated by the blue bar and scaled by the top x axis) and also identifies the cumulative ratio of the 10 most important predictors as 62.5 (i.e., together they explain 62.5% of the model, as indicated by the blue curved line and scaled by the bottom x axis). RF, Random Forest; AUC, Area Under the Curve; nMCC, normalized Matthews Correlation Coefficient; APOE, Apolipoprotein E.; SMMSE, Standardized Mini Mental State Exam; MoPaRDS, Montreal Parkinson Risk of Dementia Scale; NPI, Neuropsychiatric Inventory Questionnaire.