Table 2.

Classification accuracies and significance tests for the elastic net classifier trained on different feature combinations and trials. All shuffle tests were performed for 2500 iterations and checked for convergence.

Data tested	All trials		Trial 1		Trial 2		Trial 3
	Accuracy, %	P value	Accuracy, %	P value	Accuracy, %	P value	Accuracy, %	P value
pata (baseline)	71.0	.04	71.0	.04	71.0	.04	71.0	.04
pat-gaze-cm-conf (full)	71.0	.05	74.2	.03	54.8	.24	61.3	.17
cmb	67.7	.08	67.7	.09	6.5	.998	35.5	.81
confc	64.5	.16	58.1	.23	48.4	.63	41.9	.72
cm-conf	64.5	.13	61.3	.17	48.4	.56	41.9	.68
gazed	83.9	.002	83.9	.00	38.7	.80	61.3	.14
gaze-conf	83.9	.004	80.6	.01	51.6	.34	61.3	.17
gaze-cm	71.0	.05	64.5	.10	61.3	.14	58.1	.21
gaze-cm-conf	71.0	.05	74.2	.04	45.2	.63	51.6	.36
pat-gaze	77.4	.02	71.0	.04	67.7	.09	74.2	.03
pat-gaze-cm	71.0	.05	64.5	.11	67.7	.08	64.5	.11
pat-gaze-conf	74.2	.04	67.7	.11	35.5	.85	64.5	.14
pat-cm	58.1	.21	54.8	.25	67.7	.09	61.3	.16
pat-conf	64.5	.14	61.3	.19	64.5	.13	51.6	.35
pat-cm-conf	61.3	.17	48.4	.45	64.5	.13	32.3	.88

^apat: participant metadata.

^bcm: emotion confusion matrices.

^cconf: emotion confusion details.

^dgaze: gaze patterns.