Combining LSTM Network Ensemble via Adaptive Weighting for Improved Time Series Forecasting

Table 2

Demonstrating effectiveness of our proposed LSTM ensemble on improving forecasting performance. (a) MAE and (b) MSE.

(a)

Time series	Mean Std.	Best individual LSTM	Proposed LSTM ensemble
River Flow		0.64	0.53
Vehicles		2.22	1.85
Wine		1.09	0.94
Airline		6.43	5.58

(b)

Time series	Mean Std.	Best individual LSTM	Proposed LSTM ensemble
River Flow		1.10	0.88
Vehicles		6.25	4.45
Wine		6.78	3.65
Airline		81.73	64.11

Note: Mean±Std. indicates the average and standard deviation of MAE (or MSE) measures computed over all of the individual LSTM base models in an ensemble.