A Review on Deep Sequential Models for Forecasting Time Series Data

Table 7

A summary of published articles that used deep sequential models in different applications.

Ref.	Method	Application	Results obtained	Metrics used	Data unit
[88]	BRKGA-NN	Computer science	Generated more accurate predictions. A lot of parameters. Bad running times.	MAE = 6.47 RMSE = 6.51 MAPE = 0.01	Various public datasets
[31]	Bi-LSTM, LSTM	Forecasting the turbofan engine’s remaining service life	The model performs better at predicting RUL	RMSE, MSE, MAE, MAPE	Sensor data
[89]	DIDR-LSTM	RUL prediction	DIDR-LSTM allows more precise cross-domain RUL prediction	RMSE	Sensor data
[90]	Deep-TCNN	Electricity, traffic, and spare parts of cars demand	The framework outperformed existing state-of-the-art approaches.	NRMSE, SMAPE, MASE	Hourly and monthly
[91]	TCNN	Speech enhancement	Significantly outperforms existing real-time systems in the frequency domain.	STOI and PESQ scores	Hourly
[92]	TCNN	Computer science	It was suggested that a technique for deciding on scalability be developed.	MSE	-
[93]	Hybrid ES-LSTM	M4 competition	The ES retains key elements like seasonality. The prediction of forex price LSTM networks allows for nonlinear trends and cross-learning from various related series.	MAPE = 0.45–0.49	Monthly and hourly
[94]	STCNN	Speech enhancement	STCNN architecture outperformed comparable neural network models	-	Hourly
[95]	ARIMA-LSTM, ARIMA-LSTM-DP	Well production	Hybrid ARIMA-LSTM-DEEP model is more reliable.	RMSE, MAE, MAPE, Sim	Daily
[96]	SARIMA-CNN_LSTM	Tourism demand	SARIMA-CNN-LSTM model beats other individual models.	RMSE, MAPE	Daily
[97]	Pi-sigma ANN	Computer science	The FTS-ARMA TPS-ANN, which is based on both ARMA and PSO algorithms, considerably improves prediction performance.	MAPE, RMSE	Various time series data
[98]	Pi-sigma ANNs + fuzzy c-means	Temperature and stock exchange forecasting	The best forecasts were generated by our suggested technique for all test data of each dataset	RMSE, MAE	Daily