Classification of behaviour in housed dairy cows using an accelerometer-based activity monitoring system

Table 3 Performance of the decision-tree classification algorithm across individual cows

Cow	Lying		Standing		Feeding
Cow	Sensitivity	Precision	Sensitivity	Precision	Sensitivity	Precision
1a	91.59	89.91	80.65	84.75	91.30	93.33
1b	69.62	98.21	No obs	0	100	75.38
2	71.17	94.05	68.75	68.75	92.54	92.54
3a	21.82	100	100	27.59	100	27.59
3b	60.27	100	91.67	40.74	98.61	81.61
4a	100	95	95	100	93.75	100
4b	85.59	100	100	23.26	100	71.05
5a	66.67	100	0.00	0	88.89	53.33
5b	47.83	100	39.39	15.85	78.49	51.41
6a	85.00	100	100	85.19	100	84.62
6b	100	94.12	89.04	85.53	84.62	80.00

Results were obtained using a 1-min window and values of 0.0413 and -0.055 g for threshold A and B, respectively. Characters a (day 1) and b (day 2) after the cow number correspond to the day the observations were recorded. For all of the cows, observations were recorded over two successive days. For cow number 2, no accelerometer records were collected in the second day due to a battery failure in the sensor. No observations of standing behaviour were recorded for cow 1 on day 2