Forecasting Tourism Data Using Neural Networks - Multiscale Autoregressive Model
Abstract: Neural Networks -
Multiscale Autoregressive (NN-MAR) model is a development of neural network.
The network is built by using wavelet theories for time series forecasting.
There are few research explaining how NN-MAR model can be used for forecasting
seasonal time series data. The main aspect for forecasting seasonal time series
data is the lag inputs, which should include seasonal lags. The procedure
starts from Maximal Overlap Discrete Wavelet Transform (MODWT) decomposition.
For non-stationary data, the differencing process is used to get a stationary data.
From the decomposition process we get the scale and wavelet coefficients. The
lags of these coefficients are used as the inputs in the network. In the hidden
layer, the number of hidden neurons is chosen by using the criterion of R2incremental
and F-test, so that we get the best NN-MAR model. The aim of this research is
to build NN-MAR model for seasonal time series data, such as tourism data. The
number of international tourist coming to Soekarno-Hatta airport in Jakarta and
to Ngurah Rai airport in Bali are used as the case study.
Keywords: Neural networks,
Multiscale, MODWT, NN-MAR, Tourism, Time series
Author: Brodjol Sutijo,
Suhartono dan Alfonsus J. Endharta
Journal Code: jpfisikagg110004
