Table 8.

Accuracy (ACC), precision (PPV, Positive Predictive Value) and sensitivity (TPR, True Positive Rate) for each functional fitness movement, related to the cubic kernel SVM algorithm with a 600 ms length windows and 10% overlap.

		C&J	AS	BJ	BP	TRANS
W+A	ACC	96.4	98.4	98.8	98.0	92.5
	PPV	83.5	91.8	89.1	84.8	93.7
	TPR	79.5	74.4	$76.8$	80.6	$96.5$
W+L	ACC	96.6	$98.5$	$98.7$	$97.8$	92.5
	PPV	86.1	91.3	$84.5$	$84.6$	93.6
	TPR	77.7	$77.4$	80.8	77.5	96.6
W+T	ACC	$97.3$	98.1	98.7	97.9	$93.0$
	PPV	87.8	90.4	87.7	88.4	93.7
	TPR	$84.7$	69.6	77.3	$75.1$	97.3
W+UA	ACC	97.0	$97.8$	99.1	98.0	$92.4$
	PPV	$89.5$	$88.1$	90.9	87.6	93.0
	TPR	79.0	$65.6$	83.3	78.2	97.2
UA+A	ACC	$95.9$	98.2	99.1	$98.6$	92.6
	PPV	$81.2$	90.6	$92.8$	89.1	93.8
	TPR	$75.4$	71.1	82.3	$87.7$	96.6
UA+L	ACC	96.7	98.4	99.0	98.5	$93.3$
	PPV	88.6	$93.3$	89.0	89.9	93.8
	TPR	76.3	72.6	83.7	84.9	$97.6$
UA+T	ACC	96.7	98.2	$99.1$	98.5	$93.3$
	PPV	84.6	89.2	90.6	$91.6$	$94.4$
	TPR	80.9	73.7	$85.7$	83.7	96.9

Data from pairs of IMUs were used as input. Classification performance is reported as %. For sake of clarity, only results from the most stringent validation (Stage 2 = “leave one subject out” on workout dataset) are reported. Green bold numbers = best performance; red bold numbers = worst performance. W= Wrist, UP= Upper Arm, T= Thigh, A= Ankle, L= Lumbar Segment.