Longer Time Span Air Pollution Prediction: The Attention and Autoencoder Hybrid Learning Model

Table 4

Experiment on the dataset of Wulumuqi with time span of 48, node size of 32, and data segmentation of [24, 72].

Model	MAE	MAPE	COR
LSTM	12.57	90.21	0.33
GRU	11.79	77.83	0.35
SEQ2SEQ	15.02	111.47	0.32
Attention	13.91	99.75	0.21
A&A learning	10.41	67.20	0.35