Evolving multi-label fuzzy classifier with advanced robustness respecting human uncertainty

Multi-label classification has attracted much attention in the machine learning community to address problem of assigning single samples more than one class at same time. We propose an evolving multi-label fuzzy classifier (EFC-ML-FWU) which is able self-adapt and self-evolve its structure consequent parameters form multiple hyper-planes with new incoming incremental, single-pass manner especially addresses intrinsic curse dimensionality as well human label uncertainty problems, often apparent ensure advanced robustness learned parameters. The former achieved by integrating feature weights into process, specifically designed for online problems incremental manner, measuring impact features respect their discriminatory power. are integrated (i) rule evolution criterion, leading a shrinkage distances along unimportant dimensions, reduces likelihood unnecessary thus decreases over-fitting due dimensionality, (ii) first part variable-regularized RFWLS approach realized through coordinate descent algorithm, (iii) second employing correlation-based preservation using weight-based thresholds (extending classical Lipschitz constant-based threshold) within soft operations optimize feature-based weighted L1-norm on Uncertainty labels handled integration sample weights, where lower indicate higher carried sample. This leads updating update antecedent space clustering specific exploring single-label view uncertainty. Our was evaluated several data sets from MULAN repository showed significantly improved accuracy average precision trend lines compared alternative (evolving) one-versus-rest or chaining concepts, native EFC-ML method without handling performance gains up 17% AUC trends. Furthermore, interesting insights case wrong low user experience levels certainties but potentially correct were obtained.