Table 5.

Recognition performance of different FER models on the test set.

Base Model	Model No.	Pre-Trained	Loss Function	Accuracy	F1-Score	Kappa	Alpha	AUCPR	AUC
SE-ResNet-50	1	No	Softmax	50.65	46.87	43.60	42.60	62.87	90.67
	2	No	Center with softmax	46.07	39.24	38.37	36.89	55.96	87.92
	3	No	Weighted-softmax	56.37	56.41	50.14	50.09	62.27	90.49
	4	No	Center with weighted-softmax	56.90	57.06	50.74	50.71	62.24	90.33
	5	No	Weighted-cluster with weighted-softmax	56.27	56.42	50.03	49.97	61.57	90.10
	6	Yes	Softmax	52.22	49.51	45.4	44.54	63.27	90.75
	7	Yes	Center with softmax	47.08	40.02	39.51	38.27	54.91	86.75
	8	Yes	Weighted-softmax	59.72	59.72	53.97	53.93	66.47	91.85
	9	Yes	Center with weighted-softmax	59.60	59.50	53.83	53.82	65.35	91.21
	10	Yes	Weighted-cluster with weighted-softmax (proposed)	60.70	60.49	55.09	55.06	66.55	91.82