Table 2. Comparison of 1NN with SVM in terms of AUC (mean and std).

	SVM	1NN
Conditions applied	AUC	AUC	Difference	p-value
Standard 10-fold CV	0.977 (0.010)	0.956 (0.011)	0.022 (0.006)	<0.01
Fall type-wise CV	0.976 (0.012)	0.956 (0.013)	0.020 (0.012)	<0.01
Phone sampling at 25 Hz	0.969 (0.008)	0.946 (0.010)	0.022 (0.007)	<0.01
Phone sampling at 16.7 Hz	0.961 (0.009)	0.937 (0.010)	0.024 (0.008)	<0.01
Phone in hand bag	0.899 (0.011)	0.951 (0.007)	−0.053 (0.007)	<0.01

Different conditions are considered in each row. The first row is the standard cross-validation (CV). In the second row the CV is done by leaving out each time a different type of fall for testing. In the remaining rows, the validation sets for CV are taken under varying conditions. 1NN is trained and tested with data obtained under the same conditions, while SVM is trained with data obtained under “standard” conditions (50 Hz, phone in pocket).