Table 1.

Summary of the top seven wearable literature, where #Sen. corresponds to the number of sensors used and #Sub. corresponds to the number of subjects in the study. The system accuracy denotes the overall average accuracy when classifying all actions. The drink detection accuracy shows the accuracy for detecting the drinking action only. The same is applied for the F1-scores.

Ref.	#Sen.	Method	#Sub	System Accuracy (%)	Drinking Detection Accuracy (%)	System F1-Score (%)	Drinking Detection F1-Score (%)	Null Class
[46]	2	Binary CNN 1 LSTM 2	41	95.7	-	96.5	-	√
				81.4		85.5
[52]	1	5-class RNN 3 + LSTM 2	NA	99.6	100	99.2	100	×
[54]	1	Binary RF 4	6	97.4	97.4	96.7	95.3	√
		3-class ANN 5		98.2	99	95.3	93.3
		5-class ANN 6		97.8	98.6	87.2	90.9
[59]	1	2-stage CRF 7: 8-class	70	-	-	60	85.5	√
		3-class				81.1	93.4
[60]	1	Binary Adaboost	20	94.4		96.2	-	√
		5-class RF 4		-	-	91	95
[61]	5	9-class SVM 7	20	91.8	-	91.1	-	×
	2			89		88.4	93.4
[62]	3	3-stage SVM 7 + HMM 8	14	-	-	87.2	-	√

¹ Convolutional Neural Network, ² Long Short Term memory, ³ Recurrent Neural Network, ⁴ Random Forest, ⁵ Artificial Neural Network, ⁶ Conditional Random Fields, ⁷ Support Vector Machine, ⁸ Hidden Markov Model