A Representation-Enhanced Anomaly Detection Model With Prototype Learning and Neighborhood Rough Set.

Abstract

Anomaly detection (AD) has attracted increasing attention because of its importance in identifying unusual patterns across widely applications. Existing reconstruction-based AD methods have shown strong capability in modeling complex data distributions, but still face several limitations. First, most methods assume that the training data contain exclusively of normal samples (normals), limiting their applicability in practical scenarios. Second, redundant and irrelevant features in high-dimensional data weaken the discriminative capability of learned representations. Therefore, we propose a representation-enhanced AD model with prototype learning and neighborhood rough set (ADPLN). Specifically, a neighborhood rough set (NRS)-based feature selection strategy is employed to identify informative features for more discriminative representations. An adaptive prototype learning mechanism is introduced to adaptively represent diverse normal patterns. Furthermore, we design a sample-prototype correlation objective function to promote the learning of discriminative representations of normals and mitigate the negative impacts of anomalies. Extensive experiments on 13 real-world datasets demonstrate that ADPLN outperforms the state-of-the-art methods.