Identification of reindeer fine-scale foraging behaviour using tri-axial accelerometer data

Table 3 Performance statistics (%) of Random forests (RF), Support vector machines (SVM) and Hidden-Markov models (HMM) using time-domain features in 2-, 3- and 5-s windows (2 s, 3 s and 5 s)

		Grazing			Browsing high			Browsing low			Inactivity			Walking			Trotting			Other
	Window size	Se	Pr	Ac	Se	Pr	Ac	Se	Pr	Ac	Se	Pr	Ac	Se	Pr	Ac	Se	Pr	Ac	Se	Pr	Ac	K	Overall accuracy
RF	2 s	89	86	93	25	56	62	68	64	79	94	92	92	45	53	72	57	61	78	18	51	59	72	82
	3 s	89	86	93	28	57	64	73	70	82	94	92	93	49	57	73	53	58	76	21	61	60	75	84
	5 s	86	85	92	25	49	62	77	74	84	94	93	93	49	52	74	34	34	67	19	50	59	76	85
SVM	2 s	89	84	93	14	50	57	66	65	79	94	91	92	47	56	73	33	55	66	21	39	60	72	82
	3 s	89	86	93	9	36	54	72	68	81	94	92	93	48	59	73	22	56	61	19	43	59	74	83
	5 s	86	85	92	0	0	50	76	70	83	94	92	93	43	58	71	0	67	52	19	56	59	75	84
HMM	2 s	85	88	91	79	29	89	53	80	75	95	90	92	75	51	86	78	53	89	40	44	69	72	82
	3 s	85	88	91	68	26	83	54	80	75	95	90	92	69	39	82	79	40	89	36	40	67	72	82
	5 s	78	86	88	24	20	62	54	74	74	95	88	90	65	38	81	24	20	62	29	25	63	66	78

Behaviour-specific metrics are given as sensitivity (Se), precision (Pr), accuracy (Ac), and overall model performance are presented as overall accuracy and Cohen’s kappa (K)
Behaviours other than grazing, browsing high, browsing low, inactivity, walking, and trotting are included as “other” in model training
Highest behaviour-specific metrics for each model are presented in bold and the overall best performing model is highlighted in italic

ISSN: 2051-3933