WaveletFittingann {WaveletANN}

Wavelet-ANN Hybrid Model for Forecasting

Description

Wavelet-ANN Hybrid Model for Forecasting

Usage

WaveletFittingann(
  ts,
  Waveletlevels,
  Filter = "haar",
  boundary = "periodic",
  FastFlag = TRUE,
  nonseaslag,
  seaslag = 1,
  hidden,
  NForecast
)

Arguments

ts

Univariate time series

Waveletlevels

The level of wavelet decomposition

Filter

Wavelet filter

boundary

The boundary condition of wavelet decomposition

FastFlag

The FastFlag condition of wavelet decomposition: True or False

nonseaslag

Number of non seasonal lag

seaslag

Number of non seasonal lag

hidden

Size of the hidden layer

NForecast

The forecast horizon: A positive integer

Value

• Finalforecast - Forecasted value

• FinalPrediction - Predicted value of train data

• Accuracy - RMSE and MAPE for train data

References

• Aminghafari, M. and Poggi, J.M. 2012. Nonstationary time series forecasting using wavelets and kernel smoothing. Communications in Statistics-Theory and Methods, 41(3),485-499.

• Paul, R.K. A and Anjoy, P. 2018. Modeling fractionally integrated maximum temperature series in India in presence of structural break. Theory and Applied Climatology 134, 241–249.

Examples

N <- 100
PHI <- 0.2
THETA <- 0.1
SD <- 1
M <- 0
D <- 0.2
Seed <- 123
set.seed(Seed)
Sim.Series <- fracdiff::fracdiff.sim(n = N,ar=c(PHI),ma=c(THETA),d=D,rand.gen =rnorm,sd=SD,mu=M)
simts <- as.ts(Sim.Series$series)
WaveletForecast<-WaveletFittingann(ts=simts,Waveletlevels=3,Filter='d4',
nonseaslag=5,hidden=3,NForecast=5)