Table 3.

Overall classification performance (accuracy, in %) for k-Nearest Neighbors (kNN) algorithms, as a factor of different kernel functions, window lengths and percentage of window overlap.

		Overlap [%]			Overlap [%]			Overlap [%]
		0	10	20	0	10	20	0	10	20
		Fine Class			Medium Class			Coarse Class
Window [ms]	300	96.4	96.4	96.7	96.3	96.0	96.5	89.2	89.8	90.6
	400	96.3	96.3	97.0	96.1	96.5	96.8	$89.0$	90.1	90.5
	500	96.1	97.0	97.0	96.0	96.5	96.1	89.7	89.8	90.4
	600	96.2	$97.2$	96.7	95.7	96.4	96.2	$89.0$	89.9	90.5
		Cosine			Cubic			Weighted
Window [ms]	300	96.4	96.3	96.6	94.0	93.7	94.1	96.4	96.3	96.8
	400	96.3	96.5	96.5	93.9	94.2	94.9	96.4	96.7	96.9
	500	96.1	96.7	96.3	94.6	94.7	94.8	96.3	96.6	96.4
	600	96.1	$97.2$	96.5	94.7	94.9	94.8	96.3	96.9	96.5

Data from all the five IMUs available were used as input. Green bold numbers = best performance; red bold numbers = worst performance.