Fire-Net: A Deep Learning Framework for Active Forest Fire Detection

Table 3

Accuracy assessment of active fire detection (Comparison of five common Classification methods).

Method	OA (%)	Precision (%)	Recall (%)	FPR (%)	MD (%)	F1-score (%)	KC
(a) Australian’s Forest
MLP	99.02	0.00	0.00	0.009	100	0.00	0.00
KNN	97.13	25.13	99.63	2.88	0.37	40.14	0.391
RF	96.29	20.66	100	3.73	0.00	34.24	0.331
SVM	99.03	14.28	0.01	0.001	99.98	0.03	0.03
XGBOOST	96.29	20.66	100	3.73	0	34.24	0.331
(b) Central Africa’s Forest in 2018-12-19
MLP	99.48	0.00	0.00	0.51	100	0.00	0.1964
KNN	99.99	79.95	70.20	0.01	29.79	72.96	0.4293
RF	99.99	57.47	75.75	0.02	24.24	65.35	0.4293
SVM	99.99	0	0	0	100	0	0.4293
XGBOOST	99.99	57.47	75.75	0.02	24.24	65.35	0.4293
(c) Brazil‘s Forest
MLP	99.64	0.002	0.08	0.34	99.91	0.34	0.1426
KNN	99.99	60.16	91.77	0.006	8.22	0.006	0.4294
RF	99.98	49.93	98.19	0.011	1.80	66.20	0.4294
SVM	99.98	50.00	0.02	0.00	99.97	0.04	0.4292
XGBOOST	99.98	49.93	98.19	0.011	1.80	66.20	0.4294
(d) Chernobyl
MLP	99.99	0	0	1.27	100	0	0.2705
KNN	99.99	83.84	75.60	0.00	24.39	79.51	0.4293
RF	99.99	70.35	91.17	0.00	8.83	79.42	0.4293
SVM	99.99	0	0	0	100	0	0.4293
MLP	99.99	70.35	91.17	0.00	8.83	79.42	0.4293