Air Quality Prediction Model Based on Spatiotemporal Data Analysis and Metalearning

Table 5

RMSE of transfer learning methods over different sizes of training dataset in the target city. (taking Beijing as the target city and other cities as the source cities).

Methods		24 h	72 h	240 h	720 h	2400 h
Single-FT	Tianjin	62.7	62.1	53.5	50.5	45.6
	Shenzhen	71.3	70.7	67.6	61.0	55.2
	Guangzhou	70.9	69.0	66.2	59.8	54.5
Multi-FT		63.1	61.2	55.3	52.5	43.1
MAML		59.4	57.8	49.4	45.1	36.2
MetaGAT-LSTM		55.8	53.7	45.3	40.2	31.4