Table 7 Results for mid-term forecasting horizon (24 h); split: 90–10.

From: CO2 concentration forecasting in smart cities using a hybrid ARIMA–TFT model on multivariate time series IoT data

Model

MAE

RMSE

MAPE

Retrain

ARIMA

52.0185

63.5268

23.424

No

ARIMA

52.0185

63.5268

23.424

Yes

DeepAR

35.4227

45.8422

14.8968

Yes

DeepAR

83.7181

103.7674

37.3648

No

ES

65.3607

78.0223

29.1344

No

ES

65.3607

78.0223

29.1344

Yes

FFT

42.2596

52.1752

16.9636

No

FFT

42.2596

52.1752

16.9636

Yes

Hybrid

29.4695

36.5776

13.164

No

Hybrid

30.7673

38.0473

13.682

Yes

LSTM

78.7371

95.8902

36.6992

No

LSTM

86.3846

106.3583

39.5714

Yes

N-BEATS

55.2179

68.9673

24.7866

No

N-BEATS

65.2014

83.7045

29.6462

Yes

TCN

58.733

70.0161

24.57

No

TCN

92.241

120.2069

42.0035

Yes

TFT

31.5093

37.5722

13.7158

No

TFT

46.5937

53.5309

19.7722

Yes

Theta

64.6421

77.2401

28.9297

No

Theta

64.6421

77.2401

28.9297

Yes

Transformer

51.3384

60.9133

21.1374

No

Transformer

78.3312

98.7249

36.9034

Yes

  1. Ranked by name ascending, MAE descending. Bold means the lowest error/best performance.