Improved mutual information method in combination model selection for forecasting tourist arrival

Abstract

During the past several decades, a considerable amount of studies has been carried out on finding the highest accurate forecast model. Recently, it has been demonstrated that combining forecasts of individual models can improve forecast performance. Nevertheless, in practice, selecting individual forecast for model combination based on forecast accuracy evaluation might not have extracted all the significant information for the actual output forecast values. Hence, it is advocated to select the optimal individual model from theoretical and experimental aspects that may be able to offer more information to provide a better prediction of combination forecast model. Thus, the mutual information algorithm scaling proposed (MI-S-P) approach is proposed in this study to select the optimal individual model as an input for combination forecast model. Seven individual models and three linear combination methods are applied in this study to evaluate the effectiveness of the MI-S-P approach. The data used in this study is a short term 12 months ahead forecast which includes the monthly data on the top five international tourists arrival entering into Malaysia from the year 2000 to 2013. The results from this study is divided into two main parts, namely in-sample data (fitted model) and out-sample data (forecast model). The analyses show that the in-sample and out-sample values using MI-S-P model has successfully improve forecast accuracy on average by 2% compared to using all of individual forecast combination models. This study concludes that MI-S-P approach can be an alternative way in identifying the right optimal individual model for modelling combination forecast model.