An evolving neuro-fuzzy system based on uni-nullneurons with advanced interpretability capabilities

This paper proposes a hybrid architecture based on neural networks, fuzzy systems, and n-uninorms for solving pattern classification problems, termed as ENFS-Uni0 (short evolving neuro-fuzzy system uni-nullneurons). The model can produce knowledge in an on-line (single-pass) learning context particular form of rules representing the dependencies among input features through IF-THEN type relations. antecedents are thereby realized uni-nullneurons, which constructed from n-uninorms, leading to possibility express both, AND- OR-connections (and mixture these) single antecedent parts rule thus achieving advanced interpretability aspect rules). neurons’ evolution is done extended version autonomous data partition method (ADPA). On-line interpretation timely addressed by (i) concept tracking degree changes over stream samples, may indicate experts/operators how much dynamics process be used structural active component request operator’s feedback case significant (ii) updating feature weights incrementally. These (possibly changing) impact degrees problem: with low seen unimportant masked out when showing expert (? length reduction). rules’ consequents represented certainty vectors recursively updated indicator-based recursive weighted least squares (I-RWLS) approach (one RWLS estimator per class) where given neuron activation levels order gain stable local learning. proposed this was successfully compared related approaches literature classifying binary multi-class patterns. results obtained show outperformance works terms higher accuracy trend lines time, while offering high coherent solve problems.