Fig 1.
A normal setup of a training process for Machine Learning and Deep Learning frameworks.
Table 1.
The summary of six existing datasets for the air quality prediction problems.
Table 2.
The architectures of SAITS, BRITS, MRNN, and transformer models.
Fig 2.
The experimental results on Frankfurt dataset.
Fig 3.
The experimental results on Beijing dataset.
Fig 4.
The experimental results on Northern Taiwan dataset.
Fig 5.
The experimental results in Dalat dataset.
Due to the high missing rates in the original Dalat dataset greater than 40%, we generate extra artificial missing values with missing rates of 10%–30% only.
Fig 6.
The experimental results in the Cau Giay District dataset.
Fig 7.
The experimental results on Minh Khai District dataset.
Table 3.
Experimental results of different methods in Frankfurt air quality datasets.
Table 4.
Experimental results of different methods in Beijing air quality dataset.
Table 5.
Experimental results of various methods on Northern Taiwan air quality dataset.
Table 6.
Experimental results of different methods on Dalat (Vietnam) air quality dataset.
Table 7.
Experimental results of different methods on Cau Giay (Vietnam) air quality dataset.
Table 8.
Experimental results of different methods on Minh Khai District (Vietnam) air quality dataset.
Fig 8.
Evaluating MAE of the SAITS method with the different number of layers on various datasets.
Fig 9.
Evaluating RMSE of the SAITS method with the different number of layers on various datasets.
Fig 10.
Evaluating the running time of the SAITS method with the different number of layers on various datasets.