Table 3.

Comparison between critical slowing model (Maturana et al., 2020) and Paradigm 2.

	Critical slowing		LSTM		LSTM with matched sensitivity		LSTM with matched time in high
	Sensitivity	Time in High	Sensitivity	Time in High	Sensitivity	Time in High	Sensitivity	Time in High
Patient 1	0.83	0.08	0.22	0.02	0.83	0.42	0.48	0.08
Patient 2	0.87	0.0002	0.69	0.004	0.85	0.38	0.61	0.001
Patient 3			0.49	0.12
Patient 6	0.66	0.03	0.28	0.06	0.66	0.21	0.08	0.03
Patient 8	0.64	0.23	0.78	0.20	0.64	0.12	0.81	0.23
Patient 9	0.85	0.16	0.84	0.26	0.85	0.26	0.72	0.16
Patient 10	0.78	0.24	0.58	0.17	0.78	0.34	0.64	0.24
Patient 11	0.86	0.16	0.38	0.04	0.86	0.26	0.72	0.16
Patient 13	0.64	0.14	0.75	0.17	0.64	0.10	0.72	0.14
Patient 15	0.87	0.0007	0.08	0.01	0.83	0.73	0.03	0.0007

Compares the results of the critical slowing model (Maturana et al., 2020) with the LSTM model for paradigm 2. The columns under the LSTM heading represent the sensitivity and time in high for when the pre-ictal label was higher than the inter-ictal label or a threshold of 0.5 on the ROC curve. Matching the sensitivity or time in high to the published values in the critical slowing paper allows a direct comparison between the results. When the current LSTM model performed better than critical slowing model, values are highlighted in bold.