[Retracted] An Economic Forecasting Method Based on the LightGBM-Optimized LSTM and Time-Series Model

Table 1

Combination of different activation functions and optimization algorithms.

Activation function	Optimization algorithm	Closing price
		RMSE		Mae
		Training error	Test error	Training error	Test error
ReLU	RMSprop	0.01278	0.01272	0.00976	0.00970
ReLU	Adagrad	0.01307	0.01367	0.00904	0.00867
ReLU	Adadelta	0.01470	0.01529	0.01013	0.00968
ReLU	Adam	0.01082	0.01049	0.00726	0.00647
ReLU	Adamax	0.01080	0.01107	0.00711	0.00692
Sigmiod	RMSprop	0.01299	0.01464	0.00889	0.01168
Sigmiod	Adagrad	0.01566	0.01793	0.01107	0.01368
Sigmiod	Adadelta	0.02534	0.02383	0.02193	0.02047
Sigmiod	Adam	0.01188	0.01083	0.00899	0.00693
Sigmiod	Adamax	0.01320	0.01322	0.01005	0.01002
Tanh	RMSprop	0.01176	0.01054	0.00834	0.00658
Tanh	Adagrad	0.01253	0.01314	0.00855	0.00835
Tanh	Adadelta	0.01539	0.01666	0.01050	0.01089
Tanh	Adam	0.01056	0.01108	0.00686	0.00743
Tanh	Adamax	0.01069	0.01075	0.00702	0.00653
Softsign	RMSprop	0.01108	0.01076	0.00770	0.00651
Softsign	Adagrad	0.01279	0.01355	0.00866	0.00869
Softsign	Adadelta	0.01500	0.01532	0.01052	0.00994
Softsign	Adam	0.01104	0.01166	0.00711	0.00772
Softsign	Adamax	0.01078	0.01110	0.00684	0.00664
Softplus	RMSprop	0.02830	0.03456	0.02395	0.03233
Softplus	Adagrad	0.01453	0.01534	0.00991	0.00975
Softplus	Adadelta	0.04457	0.04699	0.04106	0.04471
Softplus	Adam	0.01337	0.01476	0.01000	0.01216
Softplus	Adamax	0.01165	0.01214	0.00813	0.00871