A Review on Deep Sequential Models for Forecasting Time Series Data

Table 3

A summary of published articles that used deep sequential models in environment and pollution predictions.

Ref.	Method	Application	Results obtained	Metrics used	Data unit
[61]	Hybrid ANN + CSA	Environment pollution	The model was consistently outperforming control models.	Error = 24.1%	Monthly
[62]	Hybrid CEEMDAN-deep-TCNN	Environment pollution	The model shown to perform the best compared to time series models, ANN, and popular deep learning models.	MAPE, RMSE, MAE	Hourly
[63]	LSTM	Environment pollution	LSTM outperformed back-propagation in terms of prediction accuracy.	RMSE = 0.08	Air pressure parameters
[64]	LSTM	Environment pollution	There is no “one-size-fits-all” approach that is effective in any city under any circumstance.	RMS = 30–40 ppm	Hourly
[65]	LSTM	Environment pollution	The LSTM model’s root mean square error is 5.58	Average error was 5.58 by RMSE	Monthly
[66]	LSTM-BP	Environment pollution	The time series predicted by LSTM-BP was found to be more accurate	RMSE = 1.03. MAE = 0.68 MAPE = 3.37	Daily
[67]	ANN	River flow	The model significantly increases when the lag 1 river flow is included as an input.	R^2, MIA, RMSE, MNSE	Daily and monthly
[68]	Hybrid ICEEWT-IGWO-GRU	Streamflow	The proposed model outperformed the single GRU	MAE, RMSE	Monthly