Table 7.

Performance metrics of the classification algorithm. Summary of the precision, recall, F-score and specificity metrics for sheep activity classification using both the ear-mounted and collar-mounted sensors with window sizes of 3, 5 and 7 s with sampling frequencies of 8, 16 and 32 Hz. In bold are the highest values for precision, recall, F-score and specificity when comparing across window sizes. In italics are the highest values for precision, recall, F-score and specificity when comparing only across different sample frequencies.

			window size
			precision			recall			F-score			specificity
sensor position	behaviour	sampling frequency	3 s	5 s	7 s	3 s	5 s	7 s	3 s	5 s	7 s	3 s	5 s	7 s
ear	walking	8 Hz	86	90	88	80	84	83	97	97	97	83	87	86
		16 Hz	85	86	88	83	86	88	96	96	97	84	86	88
		32 Hz	92	93	91	87	92	90	99	99	98	89	93	91
	standing	8 Hz	81	85	83	82	85	86	94	96	95	82	85	84
		16 Hz	86	88	90	85	86	89	93	94	95	86	87	89
		32 Hz	92	95	93	93	92	93	96	98	97	93	93	93
	lying	8 Hz	94	94	95	96	96	96	93	93	94	95	95	95
		16 Hz	92	93	94	93	94	95	94	95	95	92	94	94
		32 Hz	96	96	97	97	98	97	96	96	97	97	97	97
collar	walking	8 Hz	81	86	87	81	83	83	95	96	97	81	85	85
		16 Hz	85	88	89	89	88	91	96	97	97	87	88	90
		32 Hz	89	94	91	90	92	91	98	99	98	90	93	91
	standing	8 Hz	81	85	83	79	84	84	94	95	95	80	84	83
		16 Hz	89	89	91	87	90	91	96	95	96	88	90	91
		32 Hz	92	93	93	92	93	93	96	97	97	92	93	93
	lying	8 Hz	95	95	95	96	97	97	95	95	94	96	96	96
		16 Hz	93	95	96	93	94	95	95	96	97	93	94	95
		32 Hz	97	97	97	97	98	98	97	97	97	97	97	97