Vision-Based Deep Q-Learning Network Models to Predict Particulate Matter Concentration Levels Using Temporal Digital Image Data

Table 4

Sensitivity (Sen), specificity (Spe) and Youden index (= sensitivity + specificity - 1) of predictive models with an application to various video sequences.

# of frames	5 frames			10 frames			15 frames			20 frames
	Konkuk Univ. Indoor
	Sen	Spe	Youden	Sen	Spe	Youden	Sen	Spe	Youden	Sen	Spe	Youden
Deep Q Haze	0.9873	0.9963	0.9836	0.9927	0.9963	0.9890	0.9927	0.9967	0.9894	0.9890	0.9927	0.9817
Deep Haze	0.9575	0.4650	0.4225	0.9850	0.4300	0.4150	0.9750	0.5150	0.4900	0.9800	0.5175	0.4975
RF	0.9850	0.2675	0.2525	0.9825	0.2774	0.2599	1.0000	0.0000	0.0000	1.0000	0.0000	0.0000
SVM	1.0000	0.0000	0.0000	1.0000	0.0000	0.0000	1.0000	0.0000	0.0000	1.0000	0.0000	0.0000
	Konkuk Univ. Outdoor
Deep Q Haze	0.8550	0.9366	0.7916	0.8500	0.9500	0.8000	0.8500	0.9600	0.8100	0.8600	0.9633	0.8233
Deep Haze	0.3760	0.5360	-0.0880	0.4120	0.4880	-0.1000	0.3640	0.5400	-0.0960	0.3740	0.5420	-0.0840
RF	0.5240	0.4140	-0.0620	0.5300	0.4820	0.0120	0.4080	0.4679	-0.1241	0.4540	0.4679	-0.0781
SVM	0.7320	0.5800	0.3120	0.7180	0.5360	0.2540	0.7020	0.5460	0.2480	0.7380	0.5500	0.2880
	Keimyung Univ.
Deep Q Haze	0.9871	0.9814	0.9685	0.9885	0.9842	0.9727	0.9885	0.9871	0.9756	0.9914	0.9914	0.9828
Deep Haze	0.8760	0.4250	0.3010	0.8980	0.3950	0.2930	0.8820	0.4216	0.3036	0.8900	0.4200	0.3100
RF	0.8200	0.0683	-0.1117	0.8580	0.0900	-0.0520	0.8480	0.1030	-0.0490	0.8740	0.1116	-0.0144
SVM	0.8375	0.1185	-0.0440	0.8375	0.1585	-0.0040	0.8375	0.2000	0.0375	0.8375	0.1871	0.0246
	Mobile Phone
Deep Q Haze	0.9733	0.7874	0.7607	0.9866	0.7632	0.7498	0.9822	0.7487	0.7309	0.9777	0.7439	0.7216
Deep Haze	0.9130	0.1288	0.0418	0.9178	0.1288	0.0466	0.9130	0.1244	0.0374	0.9082	0.1244	0.0326
RF	0.7004	0.2711	-0.0285	0.7681	0.2622	0.0303	0.6908	0.2311	-0.0781	0.7198	0.2577	-0.0225
SVM	0.6280	0.0533	-0.3187	0.5990	0.0666	-0.3344	0.5990	0.0533	-0.3477	0.5893	0.0622	-0.3485