Air Pollution Concentration Forecast Method Based on the Deep Ensemble Neural Network

Table 3

The accuracy comparisons between NN, RNN, LSTM, and EN with different numbers of neurons.

	Model	The number of neurons of dense layers	MAE	MAPE
Group 1	NN	5	6.99	19.59%
	RNN	5	7.23	20.17%
	LSTM	5	6.63	20.01%
	GRU	5	6.73	20.23%
	EN1 (NN+RNN)	{5, 5}	7.19	20.31%
	EN2 (NN+LSTM)	{5, 5}	6.62	19.65%
	EN3 (RNN+LSTM)	{5, 5}	6.59	19.35%
	EN4 (NN+RNN+LSTM)	{5, 5, 5}	6.73	19.61%
	EN5 (NN+RNN+LSTM+GRU)	{5, 5, 5, 5}	6.72	19.60%
	EN (RNN+LSTM+GRU)	{5, 5, 5}	6.60	19.99%
Group 2	NN	10	6.89	19.17%
	RNN	10	6.57	18.70%
	LSTM	10	8.27	25.43%
	GRU	10	7.23	19.88%
	EN1 (NN+RNN)	{10, 10}	6.54	18.30%
	EN2 (NN+LSTM)	{10, 10}	7.02	20.76%
	EN3 (RNN+LSTM)	{10, 10}	6.84	20.68%
	EN4 (NN+RNN+LSTM)	{10, 10, 10}	6.65	19.49%
	EN5 (NN+RNN+LSTM+GRU)	{10, 10, 10, 10}	6.67	19.23%
	EN (RNN+LSTM+GRU)	{10, 10, 10}	6.35	17.36%
Group 3	NN	15	7.11	19.89%
	RNN	15	6.57	16.77%
	LSTM	15	6.45	17.34%
	GRU	15	6.56	19.84%
	EN1 (NN+RNN)	{15, 15}	6.42	17.68%
	EN2 (NN+LSTM)	{15, 15}	6.49	17.06%
	EN3 (RNN+LSTM)	{15, 15}	6.22	16.25%
	EN4 (NN+RNN+LSTM)	{15, 15, 15}	6.26	16.64%
	EN5 (NN+RNN+LSTM+GRU)	{15, 15, 15, 15}	6.25	16.56%
	EN (RNN+LSTM+GRU)	{15, 15, 15}	6.19	16.20%
Group 4	NN	20	6.95	19.19%
	RNN	20	7.23	18.49%
	LSTM	20	6.48	16.79%
	GRU	20	6.60	17.25%
	EN1 (NN+RNN)	{20, 20}	6.57	17.39%
	EN2 (NN+LSTM)	{20, 20}	6.20	16.42%
	EN3 (RNN+LSTM)	{20, 20}	6.46	16.57%
	EN4 (NN+RNN+LSTM)	{20, 20, 20}	6.24	16.28%
	EN5 (NN+RNN+LSTM+GRU)	{20, 20, 20, 20}	6.20	16.56%
	EN (RNN+LSTM+GRU)	{20, 20, 20}	6.28	16.46%
Group 5	NN	25	7.00	19.96%
	RNN	25	7.79	23.55%
	LSTM	25	7.14	21.04%
	GRU	25	7.08	19.39%
	EN1 (NN+RNN)	{25, 25}	7.10	20.87%
	EN2 (NN+LSTM)	{25, 25}	6.71	19.56%
	EN3 (RNN+LSTM)	{25, 25}	7.13	21.57%
	EN4 (NN+RNN+LSTM)	{25, 25, 25}	6.86	20.37%
	EN5 (NN+RNN+LSTM+GRU)	{25, 25, 25, 25}	6.88	20.22%
	EN (RNN+LSTM+GRU)	{25, 25, 25}	6.80	19.78%
Group 6	NN	30	7.21	20.29%
	RNN	30	7.02	19.74%
	LSTM	30	6.91	18.65%
	GRU	30	7.02	19.56%
	EN1 (NN+RNN)	{30, 30}	6.90	19.41%
	EN2 (NN+LSTM)	{30, 30}	6.66	18.51%
	EN3 (RNN+LSTM)	{30, 30}	6.57	18.34%
	EN4 (NN+RNN+LSTM)	{30, 30, 30}	6.60	18.48%
	EN5 (NN+RNN+LSTM+GRU)	{30, 30, 30, 30}	6.56	18.23%
	EN (RNN+LSTM+GRU)	{30, 30, 30}	6.47	17.80%