report.md

full_mn10_as_2s_seed42_group-source

model=mn10_as head=mlp sr=32000 clip=2.0s epochs=20 folds=5 seed=42

Held-out test metrics by fold

fold	test_acc	test_balanced_acc	test_macro_precision	test_macro_recall	test_macro_f1
1	0.9709	0.9682	0.9683	0.9682	0.9677
2	0.9909	0.9903	0.9906	0.9903	0.9904
3	0.9836	0.9817	0.9830	0.9817	0.9819
4	0.9800	0.9794	0.9778	0.9794	0.9786
5	0.9617	0.9618	0.9594	0.9618	0.9592

Mean / std across folds

metric	mean	std	min	max
test_acc	0.9774	0.0113	0.9617	0.9909
test_balanced_acc	0.9763	0.0113	0.9618	0.9903
test_macro_precision	0.9758	0.0123	0.9594	0.9906
test_macro_recall	0.9763	0.0113	0.9618	0.9903
test_macro_f1	0.9756	0.0122	0.9592	0.9904

Best fold (fold 2, by test_macro_recall) — per-class

label	precision	recall	f1	support
baby_cry	0.9954	0.9954	0.9954	216.0000
bicycle	1.0000	1.0000	1.0000	98.0000
glass_break	1.0000	0.9741	0.9869	116.0000
gunshot	0.9672	0.9916	0.9793	119.0000
macro avg	0.9906	0.9903	0.9904	549.0000
weighted avg	0.9911	0.9909	0.9909	549.0000

Best fold (fold 2) — confusion matrix

행=실제(true), 열=예측(pred)

true \ pred	baby_cry	bicycle	glass_break	gunshot
baby_cry	215	0	0	1
bicycle	0	98	0	0
glass_break	0	0	113	3
gunshot	1	0	0	118

Per-class recall across all folds

fold	baby_cry	bicycle	glass_break	gunshot
fold_01	0.9861	0.9898	0.9138	0.9832
fold_02	0.9954	1.0000	0.9741	0.9916
fold_03	0.9954	1.0000	0.9397	0.9916
fold_04	0.9861	1.0000	0.9569	0.9748
fold_05	0.9676	1.0000	0.8966	0.9832
mean	0.9861	0.9980	0.9362	0.9849
std	0.0113	0.0046	0.0314	0.0070

Pooled confusion matrix (all 5 folds, n=2745)

행=실제(true), 열=예측(pred)

true \ pred	baby_cry	bicycle	glass_break	gunshot
baby_cry	1065	0	4	11
bicycle	0	489	1	0
glass_break	0	1	543	36
gunshot	4	0	5	586