Robust Online Regression via an Interval Type-2 Fuzzy Resilient Learning Machine

Traditional online learning algorithms often struggle with noise and outliers. This paper proposes IT2-FRLM, a robust model that addresses these challenges through three key techniques: 1) unsupervised clustering to determine nonlinear antecedent parameters without relying on output targets, reducing sensitivity to output noise; 2) interval type-2 fuzzy logic for improved uncertainty handling and noise tolerance; and 3) a novel online resilient learning mechanism for robust adaptation to noisy, dynamic data streams. IT2-FRLM supports one-pass learning and incremental updates, enabling efficient processing with reduced computational demands. Experiments on synthetic and real-world datasets demonstrate its reliability and effectiveness, highlighting its potential for tasks such as time series forecasting and dynamic nonlinear modeling under low-quality data.