Airport Arrival Flow Prediction considering Meteorological Factors Based on Deep-Learning Methods

Table 3

Comparison of performances for different methods.

Prediction time horizon (min)	Incorporated features	Model	MAE	RMSE	MAPE^a (%)
30	Historic and scheduled flights	BP	1.623	2.188	16.657
		SVM	1.825	2.489	23.972
		GBRT	1.628	2.179	16.132
		LSTM	1.580	2.107	14.839
		XGBoost	1.607	2.176	15.050
		LSTM-XGBoost	1.634	2.286	15.005
	Historic and scheduled flights and meteorological variables	BP	1.594	2.148	16.243
		SVM	1.668	2.211	23.474
		GBRT	1.532	2.047	15.257
		LSTM	1.557	2.095	14.515
		XGBoost	1.511	2.023	15.036
		LSTM-XGBoost	1.443	1.989	14.735
60	Historic and scheduled flights	BP	2.408	3.457	13.447
		SVM	2.611	3.714	16.183
		GBRT	2.334	3.301	11.424
		LSTM	2.406	3.254	15.279
		XGBoost	2.307	3.301	11.071
		LSTM-XGBoost	2.319	3.331	11.730
	Historic and scheduled flights and meteorological variables	BP	2.347	3.311	11.557
		SVM	2.447	3.439	14.829
		GBRT	2.324	3.235	11.279
		LSTM	2.365	3.262	17.672
		XGBoost	2.191	3.054	10.783
		LSTM-XGBoost	2.065	2.934	10.834
120	Historic and scheduled flights	BP	3.299	5.171	9.038
		SVM	3.336	5.491	9.372
		GBRT	3.230	4.940	8.693
		LSTM	3.352	5.376	9.359
		XGBoost	3.128	4.933	8.398
		LSTM-XGBoost	3.330	5.387	8.917
	Historic and scheduled flights and meteorological variables	BP	3.166	5.070	8.452
		SVM	3.160	5.187	9.146
		GBRT	3.051	4.782	8.242
		LSTM	3.275	4.751	8.630
		XGBoost	3.039	4.734	8.031
		LSTM-XGBoost	2.889	4.591	7.811

^aMAPE covers the top 50% highest arrival flow samples in the test dataset.