Table 4.

Accuracy (%) and F1 Score of the individual RSI item classification on Caltech ADOS video datasets.

Scenario	B1		B2		B3		B4		B5
	Accu.	F1	Accu.	F1	Accu.	F1	Accu.	F1	Accu.	F1
5	83.33	0.81	38.33	0.36	83.33	0.81	61.94	0.55	42.78	0.40
6	85.56	0.79	43.06	0.41	85.56	0.79	67.22	0.64	50.28	0.50
7	83.93	0.80	52.82	0.53	83.93	0.80	72.42	0.66	50.67	0.46
11	87.78	0.82	44.48	0.44	87.78	0.82	65.95	0.61	51.79	0.53
12	85.83	0.82	40.83	0.40	85.83	0.82	64.17	0.56	54.44	0.53
13	87.70	0.82	51.98	0.51	87.70	0.82	61.15	0.57	51.98	0.49
14	87.22	0.81	34.33	0.27	87.22	0.81	68.17	0.66	41.63	0.42
Top3	83.06	0.78	45.00	0.45	83.06	0.78	68.89	0.65	52.50	0.53
Top5	85.28	0.81	45.28	0.43	85.28	0.81	61.94	0.58	50.28	0.52
All	85.28	0.81	45.28	0.45	85.28	0.81	59.44	0.56	50.00	0.51
Scenario	B6		B7		B8		B9		B10
	Accu.	F1	Accu.	F1	Accu.	F1	Accu.	F1	Accu.	F1
5	56.39	0.55	56.94	0.56	36.11	0.33	63.33	0.62	55.00	0.55
6	47.22	0.46	63.89	0.62	45.28	0.42	63.61	0.60	58.61	0.58
7	34.60	0.31	73.85	0.70	34.05	0.33	63.85	0.58	52.18	0.52
11	45.71	0.46	68.53	0.65	41.63	0.39	61.51	0.57	59.96	0.56
12	54.44	0.54	61.39	0.59	49.44	0.46	64.44	0.61	52.22	0.51
13	42.94	0.43	69.05	0.67	45.52	0.42	74.76	0.68	47.34	0.43
14	45.79	0.45	67.34	0.65	58.10	0.56	65.04	0.59	50.32	0.48
Top3	42.78	0.41	66.39	0.65	49.44	0.47	64.44	0.61	51.94	0.51
Top5	52.50	0.53	66.67	0.64	47.50	0.47	68.89	0.66	47.50	0.47
All	54.72	0.55	59.67	0.58	40.56	0.40	61.67	0.60	52.20	0.51
Scenario		B11			B12			B13
		Accu.		F1	Accu.		F1	Accu.		F1
5		61.67		0.58	38.33		0.38	33.89		0.35
6		59.44		0.56	56.94		0.58	52.22		0.52
7		52.18		0.50	52.90		0.52	42.54		0.41
11		63.45		0.58	46.43		0.47	50.71		0.49
12		59.44		0.57	43.06		0.41	37.50		0.38
13		65.60		0.58	36.35		0.35	49.29		0.48
14		59.29		0.53	43.29		0.41	43.69		0.41
Top3		66.94		0.63	47.50		0.48	56.67		0.57
Top5		66.94		0.63	45.28		0.45	54.72		0.55
All		57.22		0.53	42.78		0.43	47.78		0.48

“Top3” includes the scenario 6 (social), 11 (daily living), and 13 (loneliness), and “Top5” additionally includes scenario 5 (work/school) and scenario 14 (plans/hopes). The best results are highlighted by bold.