A New Adaptive Fuzzy Cognitive Map-Based Forecasting Model for Time Series

In a fuzzy cognitive map-based forecasting model, causal relationships (represented with a weight matrix) are constant. This may hinder the applicability of such a model. In this paper, we propose an adaptive fuzzy cognitive map-based forecasting model. Different from the existing models, the proposed model is made of a collection of fuzzy cognitive maps. Maps are constructed according to the clustering results of the so-called premises covering an entire time series. Subsequently, we use an optimization algorithm to train parameters of each fuzzy cognitive map individually. The proposed model construction procedure allows forming fuzzy cognitive maps that more flexible and, thus, suitable for forecasting of long time series. In experimental studies on synthetic time series and real time series, the proposed model performed very well in comparison with the original fuzzy cognitive map-based forecasting model and another two forecasting models.