Table 4.

Table showing the results of cross-validation tests for the predictive power of the Bayesian logistic regression classifier.

Test	Actual diagnosis	Predicted diagnosis
		Control	Young onset Alzheimer’s disease (YOAD) patient
L-1-O	Control	0.95	0.05
	YOAD patient	0.03	0.97
L-2-O	Control	0.95	0.05
	YOAD patient	0.03	0.97
L-H-O	Control	0.94	0.06
	YOAD patient	0.04	0.96

Leave-one-out (L-1-O) tests take each individual in turn and train the classifier on all other individuals’ feature vectors. The diagnostic status of that individual is predicted by the classifier and compared to the actual diagnosis (YOAD patient vs. control). Leave-two-out (L-2-O) tests take each possible pair of one control individual and one patient, train on all other individuals, and then predict the diagnosis of the original pair. The leave-half-out (L-H-O) test takes 500 random partitions of the data, with half of each diagnostic class in each partition, trains on one partition and predicts the diagnoses of the other. The columns of the table represent predicted diagnostic classes, and the rows represent actual diagnoses.