| ARMA.LSTM {ARMALSTM} | R Documentation |
Hybrid ARMA-LSTM Model for Time Series Forecasting
Description
The linear ARMA model is fitted to the time series. The significant number of PACF values of ARMA residuals are considered as the lag. The LSTM model is fitted to the ARMA residuals setting the lag value as the time step. User needs to install keras, tensorflow and reticulate packages as the prerequisite to implement this package.
Usage
ARMA.LSTM(X, p, q, arfima = FALSE, dist.model= "ged", out.sample, LSTM.units,
ACTIVATION.function = "tanh", DROPOUT = 0.2, Optimizer ="adam", Epochs = 100,
LSTM.loss = "mse", LSTM.metrics = "mae")
Arguments
X |
A univariate time series data |
p |
Order of AR |
q |
Order of MA |
arfima |
Whether to include arfima (0<d<0.5) |
dist.model |
The distribution density to use for the innovation. The default distribution for the mean model used is "ged". Other choices can be obtained from the rugarch package. |
out.sample |
A positive integer indicating the number of periods before the last to keep for out of sample forecasting. To be considered as test data. |
LSTM.units |
Number of units in the LSTM layer |
ACTIVATION.function |
Activation function |
DROPOUT |
Dropout rate |
Optimizer |
Optimizer used for optimization of the LSTM model |
Epochs |
Number of epochs of the LSTM model |
LSTM.loss |
Loss function |
LSTM.metrics |
Metrics |
Value
ARMA.fit: Parameters of the fitted ARMA model
ARMA.fitted: Fitted values of the ARMA model
ARMA.residual: Residual values of the ARMA model
ARMA.forecast: Forecast values obtained from the ARMA model for the test data
ARMA.residual.nonlinearity.test: BDS test results for the ARMA residuals
LSTM.lag: Lag used for the LSTM model
FINAL.fitted: Fitted values of the hybrid ARMA-LSTM model
FINAL.residual: Residual values of the hybrid ARMA-LSTM model
FINAL.forecast: Forecast values obtained from the hybrid ARMA-LSTM model for the test data
ACCURACY.MATRIX: RMSE, MAE and MAPE of the train and test data
References
Box, G. E., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2015). Time series analysis: forecasting and control. John Wiley & Sons.
Granger, C. W., & Joyeux, R. (1980). An introduction to long-memory time series models and fractional differencing. Journal of time series analysis, 1(1), 15-29.
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
Rakshit, D., Paul, R. K., & Panwar, S. (2021). Asymmetric price volatility of onion in India. Indian Journal of Agricultural Economics, 76(2), 245-260.
Rakshit, D., Paul, R. K., Yeasin, M., Emam, W., Tashkandy, Y., & Chesneau, C. (2023). Modeling Asymmetric Volatility: A News Impact Curve Approach. Mathematics, 11(13), 2793.
Rakshit, D., Roy, A., Atta, K., Adhikary, S., & Vishwanath. (2022). Modeling Temporal Variation of Particulate Matter Concentration at Three Different Locations of Delhi. International Journal of Environment and Climate Change, 12(11), 1831-1839.
Examples
y<-c(5,9,1,6,4,9,7,3,5,6,1,8,6,7,3,8,6,4,7,5)
my.hybrid<-ARMA.LSTM(y, p=1, q=0, arfima=FALSE, dist.model = "ged",
out.sample=10, LSTM.units=50, ACTIVATION.function = "tanh",
DROPOUT = 0.2, Optimizer ="adam", Epochs = 10, LSTM.loss = "mse", LSTM.metrics = "mae")