A Hybrid Approach Integrating Decomposition Ensemble Forecasting With Optimal Combination Selection for Air Passenger Demand Forecasting

Decomposition ensemble forecasting of air passenger demand is of continuing interest to researchers. The most common method employed to produce ensemble forecasts for demand forecasting is a linear combination of individual single-component forecasts. Nevertheless, it remains unclear how linear combinations with different weight assessments affect the accuracy of decomposition ensemble forecasting. Besides, to strengthen the overall accuracy of decomposition ensemble forecasting, it is intriguing to investigate whether significant synergy can be created by combining forecasts yielded by individual decomposition ensemble models using different weight schemes. This paper proposed a hybrid approach using several well-known weighting schemes to measure the weights of individual single-mode forecasts. The optimal combination selection from individual decomposition ensemble models was then used to construct combined models to strengthen the accuracy of decomposition ensemble forecasting. The empirical results indicated that linear combinations with optimal weights significantly outperformed those with other considered weighting schemes. Interestingly, the combined forecasts generated by the optimal combination selection from individual decomposition ensemble models significantly outperformed the best individual ensemble forecasts. It has been found that the accuracy obtained by decomposition ensemble forecasting with equal weights was only average, and the optimal combination subsets included few such models. The proposed hybrid approach can help industry practitioners make operational plans.