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

Table 9 Mean XGBoost accuracies (mean ± standard deviation in %) for the independent ADNI test set (no information rate 55.56%)

Exclusion method	Number of training subjects excluded
(base model)	0	50	100	150	200	250
Random (-)	62.01 ±1.59	60.42 ±1.28	59.51 ±2.54	59.79 ±1.37	62.57 ±1.50	64.58 ±1.42
LOO (LR)	62.01 ±1.59	60.14 ±1.73	59.72 ±1.91	61.46 ±1.68	59.03 ±1.89	56.94 ±1.58
LOO (RF)	62.01 ±1.59	61.04 ±2.35	58.54 ±1.59	61.11 ±1.64	61.74 ±2.25	59.72 ±2.04
Data Shapley (LR)	62.01 ±1.59	64.72 ±1.58	66.88 ±1.39	67.22 ±1.48	64.65 ±1.14	64.58 ±1.20
Data Shapley (RF)	62.01 ±1.59	63.61 ±1.79	66.18 ±1.55	66.81 ±1.83	67.15 ±1.20	66.46 ±1.12

Different methods were used to identify and focus on the training subjects with the most informative data. Ten repetitions with different seeds were performed for every exclusion data set. The best results are highlighted in bold

ISSN: 1758-9193