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 5.

Average results of evaluation of cross-validation of SPP-CNN and Trad-CNN based on classification and loss function.

Classification	Architecture	Loss	TP	TN	FP	FN	Accuracy	Precision	Recall	TSS	TSS Std ^(a)	PR AUC
CMX	Trad-CNN	BCE	1310	9584	2150	545	0.8	0.38	0.7	0.52	0.02	0.52
		TSS	1369	9713	2118	502	0.81	0.39	0.73	0.55	0.06	0.57
	SPP-CNN	BCE	1401	10429	1253	464	0.87	0.56	0.76	0.65	0.16	0.68
		TSS	1374	7834	3691	497	0.69	0.31	0.74	0.42	0.06	0.43

MX	Trad-CNN	BCE	242	7622	5779	71	0.58	0.06	0.77	0.35	0.17	0.08
		TSS	177	10884	2323	93	0.82	0.08	0.67	0.5	0.09	0.11
	SPP-CNN	BCE	174	10827	1994	107	0.84	0.1	0.62	0.46	0.24	0.14
		TSS	217	9460	3614	81	0.73	0.11	0.7	0.43	0.28	0.2

Notes.

^(a)

True skill statistic standard deviation.

Underlined values highlight the best result between every models using the same classification.

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.