Evolving multi-user fuzzy classifier system with advanced explainability and interpretability aspects

Evolving classifiers and especially evolving fuzzy have been established as a prominent technique for addressing the recent demands in building an incremental online manner, based on target labels typically provided by single user. We present framework interactive multi-user classifier system with advanced explainability interpretability aspects (EFCS-MU-AEI). Multiple users may provide their label feedback which own users’ are incrementally trained learning concepts. Its classification outputs amalgamated specific ensembling scheme, respecting (i.) uncertainty class due to labeling ambiguities among (ii.) different experience levels of voting weights. A major focus thereby is concentrated purpose increase quality (consistency certainty) user (labeling) feedbacks. It show reasons why certain decisions made certainty rule coverage degrees. The deduced from most active rules, reduced length statistically-motivated instance-based feature importance level concept. Another lies extracted rules order represent understandable knowledge contained problem realize behaviors parts space (= sample groups). weighting technique, uncertainties multiple forgetting weights (for handling drifts), well set merging process proposed aim high compactness transparency rules. Our approach was evaluated visual inspection scenario. could be shown that explanations fact significantly improved behavior three terms showing higher accumulated accuracy trends. Feature integration into updates achieve transparent final essential four features describe problem. Based this description, it turned out ways, i.e. groups, lower should taught improve understanding about process.