Skip to main content
Fig. 11 | Alzheimer's Research & Therapy

Fig. 11

From: Data analysis with Shapley values for automatic subject selection in Alzheimer’s disease data sets using interpretable machine learning

Fig. 11

SHAP force plot for an ADNI sMCI subject (PTID = 123_S_4904) from the training set to explain the prediction of the RF model, trained by excluding those 134 training subjects with the smallest LR Data Shapley values. SHAP force plots show that the individual prediction (f(x) = 0.80) consists of the sum of all feature Shapley values and the average model prediction (base value = 0.3704). The length of an arrow visualized the associated Shapley value of a feature. Feature expressions with large Shapley values have strong effects on the individual prediction and are shown in the middle of SHAP force plots. Pathogenic feature expressions have Shapley values higher than zero and are visualized as red arrows. Protective expressions are visualized as blue arrows. The MRI volumetric features and the cognitive test scores were min-max-normalized

Back to article page