Bypassing the static input size of neural networks in flare forecasting by using spatial pyramid pooling

Home

All issues

Volume 695 (March 2025)

A&A, 695 (2025) A65

Full HTML

Open Access

Table B.1.

Label-wise average results of SPP-CNN and Trad-CNN evaluations using the CMX classification.

Classification	Model	Loss	Class	Label	Correctly Predicted	Wrongly Predicted	Acc^a	Acc Std^b
CMX	Trad-CNN	BCE	1	X	23	4	0.83	0.33
				M	222	18	0.92	0.03
				C	1064	523	0.67	0.04

			0	NF	7840	1020	0.89	0.02
				FX	19	14	0.36	0.41
				FM	128	140	0.48	0.2
				FC	1596	975	0.61	0.06

		TSS	1	X	18	4	0.82	0.37
				M	238	19	0.92	0.04
				C	1113	478	0.69	0.08

			0	NF	7883	1067	0.88	0.02
				FX	41	34	0.68	0.29
				FM	1621	893	0.64	0.06
				FC	167	124	0.59	0.1

	SPP-CNN	BCE	1	X	26	0	1.0	0.0
				M	291	10	0.97	0.04
				C	1084	453	0.71	0.15

			0	NF	8449	589	0.93	0.06
				FX	17	2	0.5	0.5
				FM	185	75	0.76	0.15
				FC	1777	586	0.76	0.1

		TSS	1	X	33	0	1.0	0.0
				M	228	23	0.9	0.06
				C	1112	473	0.7	0.12

			0	NF	6236	2446	0.72	0.12
				FX	7	30	0.18	0.37
				FM	107	190	0.37	0.11
				FC	1483	1023	0.59	0.17

^(a)

Accuracy

^(b)

Accuracy Standard deviation

Notes. Value underlined highlight the best result among all models and Loss function.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.