Table 4 Normalized accuracy (RRMSE, MAPE), agreement indices (WI, \(\:{E}_{NS}\)), skill score (SS), and systematic bias (APB) for MA-NODE performance.

From: Continuous-time air pollutant forecasting using multi-timescale attention neural ordinary differential equations (MA-NODE)

Pollutant

Step

RRMSE (%)

MAPE (%)

WI

\(\:{\varvec{E}}_{\varvec{N}\varvec{S}}\)

SS

APB (%)

O₃ (ppb)

1

8.64

8.37

0.9916

0.9669

0.8232

1.45

 

2

9.55

9.64

0.9896

0.9596

0.7843

1.14

 

3

19.61

19.18

0.9546

0.8296

0.0945

3.01

CO (ppm)

1

7.85

6.13

0.9828

0.9346

0.8616

0.63

 

2

9.70

7.11

0.9734

0.9000

0.7884

2.41

 

3

21.00

16.51

0.8380

0.5307

0.0084

3.69

NO₂ (ppb)

1

6.12

4.98

0.9842

0.9393

0.8822

0.13

 

2

7.06

5.63

0.9781

0.9192

0.8432

0.60

 

3

16.98

14.15

0.8322

0.5326

0.0927

2.94

SO₂ (ppb)

1

12.22

7.43

0.9712

0.8963

0.8021

1.61

 

2

15.14

9.15

0.9528

0.8408

0.6962

3.45

 

3

26.51

16.54

0.8215

0.5125

0.0684

6.84

PM₁₀ (µg/m³)

1

9.84

7.44

0.9813

0.9304

0.8750

0.42

 

2

11.96

8.33

0.9710

0.8971

0.8154

1.79

 

3

27.16

21.17

0.7978

0.4695

0.0485

4.07

PM₂.₅ (µg/m³)

1

9.87

7.09

0.9863

0.9482

0.8780

1.00

 

2

12.72

8.46

0.9764

0.9138

0.7971

0.78

 

3

28.62

20.33

0.8433

0.5636

−0.0276

4.33

Back to article page