Hybrid approach based on cognitive mapping and regression analysis for forecasting in complex weakly formalized systems

Abstract

The article presents an original hybrid approach that integrates fuzzy cognitive maps and regression analysis for forecasting in weakly formalized systems characterized by high levels of uncertainty and complex, unstructured interrelationships among variables. The core idea of the approach lies in utilizing expert judgments, expressed through linguistic variables and fuzzy numbers, to adapt the weighting coefficients of the regression model. The weights obtained from cognitive analysis are incorporated into the procedure of multiparametric linear regression using the weighted least squares method. This integration enhances both the accuracy of forecasts and the interpretability of the model. The results of an empirical study conducted in the  statistical environment demonstrate that the proposed approach outperforms (  = 10.21%) not only classical single forecasting methods, such as linear regression (  = 13.80%) and neural network models, e.g., the multilayer perceptron (  = 31.24%), but also successfully competes with ensemble methods, including random forest (  = 12.35%) and gradient boosting (  = 7.75%). The proposed approach can be effectively applied to forecasting natural gas supply volumes to China, as well as to solving other tasks that require the integration of qualitative and quantitative data.