Meta-Learning Enhanced Trade Forecasting: A Neural Framework Leveraging Efficient Multicommodity STL Decomposition

Table 3

Performance comparison of various models for imported commodity value series.

Model	MAE	RMSE	MAPE
LastValuePredictor	3.01	5.25	0.2011
ARIMA	2.71	3.23	0.1921
VAR	10.18	11.79	0.6878
Bagging	2.97	2.39	0.1555
LSTM	2.31	3.70	0.1874
GRU	3.02	5.10	0.2272
DeepAR	2.62	4.38	0.2020
DeepVAR	2.96	5.26	0.2063
N-Beats	2.04	3.54	0.1705
N-Hits	1.94	3.43	0.1641
TFT	2.07	3.80	0.1628
TFSTL
Meta-TFSTL (fine-tuning)

Bold: best; underline: second best; italics: best in baseline.