Fuzzy Cognitive Maps Employing ARIMA Components for Time Series Forecasting

Abstract

In this paper, we address some shortcomings of Fuzzy Cognitive Maps (FCMs) in the context of time series prediction. The transparent and comprehensive nature of FCMs provides several advantages that are appreciated for decision-maker. In spite of this fact, FCMs also have some features that are hard to match with time series prediction, resulting in a prediction power that is probably not as extensive as other techniques can boast. By introducing some ideas from ARIMA models, this paper aims at overcoming some of these concerns. The proposed model is evaluated on a real-world case study, captured in a dataset of crime registrations in the Belgian province of Antwerp. The results have shown that our proposal is capable of predicting multiple steps ahead in an entire system of fluctuating time series. However, these enhancements come at the cost of a lower prediction accuracy and less transparency than standard FCM models can achieve. Therefore, further research is required to provide a comprehensive solution.