New RVFL Network Boosts Robustness Against Noisy Data

Random Vector Functional Link (RVFL) networks are known for their computational efficiency and strong generalization capabilities, making them attractive for various machine learning tasks. However, their performance can be significantly degraded by common real-world data issues such as noisy labels, outliers, and imbalanced datasets. This sensitivity limits their practical applicability in many scenarios. To address these limitations, a new model called the Kernel Risk-Sensitive Mean p-power based RVFL (KRPRVFL) has been introduced. This innovation combines the inherent efficiency of RVFL networks with the robustness of the kernel risk-sensitive mean p-power (KRP) criterion. By replacing the standard least-squares objective function with a KRP-based loss, KRPRVFL can adaptively reduce the influence of corrupted or unreliable data points during the training process, leading to improved stability and better generalization performance. Furthermore, the KRPRVFL framework incorporates a collaborative learning mechanism that enables adaptive interaction among its model components, further enhancing its resilience in complex and noisy environments. The use of kernel-induced feature mapping also allows the model to capture non-linear relationships without needing explicit hidden-layer selection, maintaining both efficiency and scalability. Extensive experiments on UCI and KEEL benchmark datasets confirm that KRPRVFL consistently surpasses existing baseline models in accuracy, robustness, and statistical significance, positioning it as a fast, scalable, and reliable solution for challenging classification tasks.