Scholarlink

<p>Balancing detection effectiveness with resource competence remains a challenge in distributed anomaly detection for wireless sensor networks (WSNs), especially when moving from local to globally coordinated models. While local detection saves energy, it can neglect broader anomalous patterns. Conversely, centralized approaches improve consistency but incur high communication costs. There is a demand for lightweight, scalable methods that enable global anomaly awareness without burdening the network. This study suggests DCESVM-DR, a distributed detection framework that forms local normal models (LNMs) at sensor nodes and aggregates them into a global normal model (GNM) at a relay node using compact statistical summaries. The framework employs a LEACH-CR-based hierarchical architecture where cluster heads forward LNMs to a relay. The relay builds the GNM via Min, Mean, or Median aggregation of model radii, which is then distributed back for consistent online detection. Dimensionality reduction and a centered ellipsoidal SVM (CESVM) are integrated to efficiently manage multivariate data. Evaluation on real sensor datasets with injected anomalies reveals that relay-based GNM construction improves detection performance over local-only models, achieving high detection rates in most configurations and accuracies around 98.9%, with a reduction in false negatives by up to 8%. In addition, the communication overhead remains low due to compact model exchanges. DCESVM-DR offers a scalable trade-off between detection consistency and communication efficiency, suitable for resource-aware WSN deployments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An enhanced anomaly detection framework for WSNs based on a centered ellipsoidal support vector machine and lightweight data aggregation

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Nurfazrina M. Zamry,
  • Anazida Zainal,
  • Murad A. Rassam,
  • Zanariah Zainudin

摘要

Balancing detection effectiveness with resource competence remains a challenge in distributed anomaly detection for wireless sensor networks (WSNs), especially when moving from local to globally coordinated models. While local detection saves energy, it can neglect broader anomalous patterns. Conversely, centralized approaches improve consistency but incur high communication costs. There is a demand for lightweight, scalable methods that enable global anomaly awareness without burdening the network. This study suggests DCESVM-DR, a distributed detection framework that forms local normal models (LNMs) at sensor nodes and aggregates them into a global normal model (GNM) at a relay node using compact statistical summaries. The framework employs a LEACH-CR-based hierarchical architecture where cluster heads forward LNMs to a relay. The relay builds the GNM via Min, Mean, or Median aggregation of model radii, which is then distributed back for consistent online detection. Dimensionality reduction and a centered ellipsoidal SVM (CESVM) are integrated to efficiently manage multivariate data. Evaluation on real sensor datasets with injected anomalies reveals that relay-based GNM construction improves detection performance over local-only models, achieving high detection rates in most configurations and accuracies around 98.9%, with a reduction in false negatives by up to 8%. In addition, the communication overhead remains low due to compact model exchanges. DCESVM-DR offers a scalable trade-off between detection consistency and communication efficiency, suitable for resource-aware WSN deployments.