Table 9 Ablation experiments for multi-step PM2.5 forecasting.

From: A hybrid deep learning air pollution prediction approach based on neighborhood selection and spatio-temporal attention

Approach

1–6 h

1–12 h

RMSE

MAE

RMSE

MAE

ResNet

19.417

13.549

22.594

15.481

STA-ConvNet

17.831

12.082

20.833

13.551

ConvLSTM

15.043

10.234

17.432

10.510

Res-ConvLSTM

13.749

9.851

14.519

9.574

STARes-ConvLSTM

12.581

8.371

12.630

8.521

KSC-ConvLSTM

11.301

7.272

11.369

7.035

Back to article page