Predicting Personal Exposure to PM_2.5 Using Different Determinants and Machine Learning Algorithms in Two Megacities, China

Table 2

Nested CV results of prediction models with different algorithms and predictors.

City	Algorithm	Model 1			Model 2			Model 3
			RMSE	MAE		RMSE	MAE		RMSE	MAE
BJ	MLR
	RF	^a	^a					^a
	SVM	^a
	XGBoost
	GBM							^a		^a
	ANN
NJ	MLR
	RF
	SVM
	XGBoost
	GBM	^a		^a		^a	^a
	ANN				^a		^a	^a	^a	^a

Note: model 1 included variables for ambient PM_2.5 from the nearest AQMS and meteorological factors. Model 2 included both variables in model 1 and variables from basic questionnaire. Model 3 included both variables in model 2 and variables from TAD. The values were from nested cross-validation. ^aStatistically different from referenced MLR method.