DEVELOPMENT OF AN ANFIS NEURO-FUZZY MODEL FOR FORECASTING WEAKLY STRUCTURED PROCESSES

Abstract

This article examines the problem of developing an ANFIS neuro-fuzzy model for forecasting weakly structured processes. The relevance of this study lies in the fact that weakly structured processes are typically characterized by uncertainty, nonlinearity, and multivariate dependence, which makes accurate forecasting by means of conventional statistical methods rather difficult. For this reason, an ANFIS model combining the advantages of artificial neural networks and fuzzy logic was proposed. The experimental data were formed as a time-series dataset based on meteorological, geophysical, and target indicators. The input space of the model was constructed using calendar features, a trend component, and lagged variables. The most informative features were selected using the mutual information regression method. The proposed model was built on the basis of a first-order Sugeno-type TSK-ANFIS architecture. Gaussian membership functions were used, rule centers were determined by the K-Means algorithm, and the output parameters were estimated using a ridge-regularized linear solution. According to the experimental results, the model achieved MAE = 0.388 and RMSE = 0.491 on the test set. The obtained results confirm that the ANFIS model is capable of adequately capturing the general dynamics of weakly structured processes and can be applied in practical forecasting tasks.