Information Flow-Based Fuzzy Cognitive Maps with Enhanced Interpretability

Abstract

<p>A Fuzzy Cognitive Map (FCM) is a graph-based tool for knowledge representation that intends to model any complex system through an interactive structure of nodes interacting with each other through causal relationships. Owing to their flexibility and inherent interpretability, FCMs have been used in various modeling and prediction tasks, particularly in situations where humans make final decisions, such as industrial anomaly detection. However, FCMs can unintentionally absorb spurious correlations presented in collected data during development, leading to poor prediction accuracy and interpretability. To address this limitation, this article proposes a novel framework for constructing FCMs based on the Liang-Kleeman Information Flow (L-K IF) analysis, a causal inference tool. The actual causal relationships are identified from the data using an automatic causal search algorithm, and these are then imposed as constraints in the FCM learning procedure to rule out spurious correlations and improve the predictive and explanatory power of the model. Numerical simulations were conducted by comparing the proposed approach with state-of-the-art FCM-based models, thereby demonstrating the promising performance of the developed FCM.</p>