Evaluation of Key Parameters Using Deep Convolutional Neural Networks for Airborne Pollution (PM10) Prediction

Table 16

Final ensemble model with full data.

Architecture	Activation function	First kernel size	Second kernel size	Number of neurons	RMSE	MAPE	IOA	R^2
MLP	Linear	—	—	10	14.9469	0.2433	0.8994	0.6789
		—	—	12
	ReLU	3	3	—
		10	4	—
CNN	Sigmoid	2	2	—
		2	3	—
		50	3	—
	Softplus	2	3	—
		20	6	—
		30	5	—

This final ensemble contains 10 architectures; from those, the 80% are of a convolution type. It is noted that the RMSE highly decreased for this ensemble, in comparison with the previous ensembles presented.