Improving Thailand tourism forecasting based on combinations of wavelet denoising schemes

According to tourist arrivals forecast, the influence of residual noise is still nontrivial problem. Wavelet denoising technique have been received considerable attention in noise removal. However, some important part of the original data may be removed along with the noise. This paper proposes Thailand tourism forecasting improvement based on combinations of wavelet denoising schemes. Various schemes of parameter combinations are experimented with the aim to properly remove undesirable noise while maintain useful information in time-series data. Thailand tourism monthly data over January 1999 to December 2013 as well as a benchmark artificial time-series data are tested; mean while different levels of additive noise are given with the purpose to evaluate the efficiency and robustness of the schemes. The denoising quality is measured by mean square error (MSE) as well as mean absolute percentage error (MAPE); whereas the improvement of forecasting performance is evaluated based on an improved forecasting error rate. High correlations between the quality of denoised data and the improvement of forecasting performance is denoted. The results show the best forecasting improvement of 33.35% and 17.07% for artificial and tourism time-series consecutively.