Akademik Veri Yönetim Sistemi
Applied Soft Computing, cilt.198, 2026 (SCI-Expanded, Scopus)
Though fuzzy cognitive maps have gained significant popularity in time series forecasting in recent years, more sophisticated methods are required for precise forecasting of non-linear and complex time series data in the environment of uncertainty and hesitation. In this research paper, we integrate hesitant fuzzy cognitive map (HFCM) and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and propose a novel method for time series forecasting. CEEMDAN offers better approach than empirical mode decomposition (EMD) and ensemble empirical mode decomposition (EEMD) in handling non-stationary datasets and breaks original time series data into intrinsic mode functions and residues. Proposed forecasting method uses an enhanced cognitive map with two-steps hesitancy in the weights of relationship between the nodes. Two-steps hesitancy in the weights of relationship between the nodes is implemented in a novel way by using Fermi function. Proposed forecasting method uses differential evolution to learn the weights of HFCM to avoid premature convergence and maintain population diversity. Proposed forecasting method is applied to eight diversified time series data of Sunspot, S&P 500, Lake Erie levels, CO2 at Mauna Loa, MCRF, RUT, Milk production, and MG Chaos. Minimum and maximum root mean square error 0.008 and 18.96 are observed in forecasting of time series data of MG Chaos and Milk production. Minimum and maximum mean absolute deviation of 0.006 and 18.521 are also observed in the same time series data of MG Chaos and Milk production. Acceptable range of evaluation parameter (δr<1)and performance parameter (PP>0)confirm accurate and unbiased forecasted outputs computed using proposed CEEMDAN and HFCM-based time series forecasting method.