<p>Class imbalance poses a significant challenge for learning-based systems, particularly when noisy samples or outliers are present. Although margin-based classifiers such as support vector machines (SVMs) exhibit relatively stable performance, they typically assign equal importance to all samples, which may lead to biased decision boundaries toward the majority class. To address these issues, this paper proposes a novel model, termed granular-ball intuitionistic fuzzy twin support vector machine with RVFL-based feature enhancement (RGBIFTSVM). The proposed method employs granular balls (GBs) to mitigate the influence of noise and outliers, and utilizes a random vector functional link (RVFL) network to enhance feature representation in a nonlinear manner. Furthermore, an intuitionistic fuzzy scoring mechanism is introduced to characterize uncertainty and emphasize minority class samples. Finally, classification is performed using a twin support vector machine (TSVM) in the transformed feature space. Extensive experiments are conducted on 19 benchmark datasets from UCI and KEEL under both clean and noisy conditions. The results demonstrate that the proposed method achieves competitive classification performance and exhibits strong robustness in handling imbalanced data.</p>