Interpreting Temporal Dynamics in Distributed Fuzzy Cognitive Maps for Time Series Prediction

Fuzzy Cognitive Maps (FCMs) are powerful tools for the modeling of complex systems, one of which is time series modeling. Although advocated for their interpretability, existing methods measure concept relevance based on static structural features often overlooking the dynamic influences over the time. This paper presents a novel SHapley Additive exPlanations (SHAP) framework based Distributed FCM model. The proposed approach invokes functions to determine the SHAP concept relevance scores for the Distributed Fuzzy Cognitive Maps (DFCMs) for time series prediction, marginalizing over each concept and comparing the predicted states to an expected baseline iteratively. This method is interpreted as measuring the time-varying impact of each concept through the dynamic course of the system. The SHAP-based approach improves the interpretability and application of FCMs to the modeling of complex time-course systems.