Incorporating a New Weighting Scheme into Decomposition Ensemble Models for Forecasting Air Passenger Flow by Combining Fuzzy Cognitive Maps with Grey Relational Analysis

Previous studies in passenger flow forecasting commonly employ decomposition ensemble models with linear addition, treating individual single-component forecasts with equal weights, to produce ensemble forecasts. It is known that fuzzy cognitive maps (FCMs) in multiple criteria decision-making are capable of modeling system dynamics by catching the causal relationships between the concepts describing a system. To enhance the forecasting ability of decomposition ensemble models with linear addition, this study aims to develop weighting schemes based on FCMs and grey relational analysis (GRA). Time series are decomposed into several components, and neural networks are applied to forecast individual components. Then, GRA is applied to assess the weights for individual single-component forecasts. To obtain ensemble forecasts, an optimal FCM determined by a genetic algorithm is used to determine the final combination weights for individual single-component forecasts. In comparison with benchmark models, the results show that the proposed FCM-based decomposition ensemble models significantly effectively improve the forecasting accuracy of air passenger flow in Taiwan across different forecasting horizons.