NK-DCHS: An adaptive hybrid immune model for imbalanced anomaly detection

Abstract

Anomaly detection remains a challenging task, especially in scenarios involving highly imbalanced data and dynamic, nonstationary patterns. Traditional machine learning methods often struggle to maintain both adaptability and interpretability in such contexts. Inspired by the adaptive nature of the human immune system, this paper presents a novel hybrid framework-Natural Killer Cell-Dendritic Cell Hybrid System (NK-DCHS)-which integrates the Sigmoid Dendritic Cell Algorithm (SDCA) with the Natural Killer Cell Algorithm (NKA) to address these limitations. The NK-DCHS framework comprises five key phases: pathogen proliferation, antigen recognition, NK cell-mediated defense, dendritic cell antigen presentation, and innate immune memory. NKA introduces adaptive local decision-making through cluster-based immune classification, overcoming the linearity limitations of traditional dendritic cell models. Meanwhile, SDCA enhances feature selection and the transformation of immunological signals (danger and safety), thereby improving decision transparency and interpretability. The combined effect of NKA and SDCA allows the model to detect anomalies with high granularity while retaining biological plausibility. Extensive experiments using 5-fold cross-validation reveal that the proposed approach consistently outperforms conventional and state-of-the-art models across multiple metrics. These results highlight the potential of NK-DCHS as a robust and adaptive solution for real-world anomaly detection tasks.