Visualization of evolving fuzzy rule-based systems

Evolving fuzzy systems are data-driven fuzzy (rule-based) systems supporting an incremental mode of model adaptation in dynamically changing environments; typically, such models are learned on a continuous stream of data in an online manner. This paper advocates the use of visualization techniques in order to help a user gain insight into the process of model evolution. More specifically, dynamic parallel coordinates, vertical parallel coordinates and rule chains are introduced as novel visualization techniques for the inspection of evolving Takagi-Sugeno-Kang (TSK) fuzzy systems. These techniques are realized in the software tool FISVis (Fuzzy Inference System Visualizer), the architecture and functionality of which are presented in this work. To show the usefulness of the proposed techniques, we illustrate their application in the context of learning from data streams with temporal concept drift.