Scholarlink

<p>The rapid growth of Internet of Things (IoT) ecosystems has created new challenges for building intelligent and privacy-preserving Intrusion Detection Systems (IDS). Traditional centralized solutions are often limited by latency, data imbalance, and exposure of sensitive information. To address these challenges, this study proposes a novel framework called Class-Aware Curriculum Federated Learning (CACFL), which integrates curriculum learning, class-dependent scheduling, and geometric median aggregation within a federated architecture. The proposed framework combines structured, temporal, and contextual features from multimodal IoT data—such as network traffic, device telemetry, and behavioral logs—to enhance intrusion detection performance. It incorporates conditional data augmentation through a CVAE-GAN model and employs an adaptive optimization strategy using the hybrid Bat–Kestrel–Antlion (BKA) optimizer to dynamically tune learning parameters. The framework leverages the complementary strengths of XGBoost, 1D-CNN, and Elastic Weight Consolidation (EWC) models for robust and generalizable feature extraction. Experimental validation demonstrates that CACFL achieves improved detection stability, resilience to non-IID data, and enhanced recognition of minority attack patterns. The study concludes that the proposed approach offers a scalable and privacy-conscious foundation for next-generation intrusion detection in large-scale IoT environments.</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

Multi-modal class-aware curriculum federated learning fusion for advanced intrusion detection using conditional generative augmentation and real-time neural optimization

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Ali Aalsaud,
  • Mohammed Fadhil Ahmed,
  • Aws Ahmed Jaff,
  • Darun Mudhafar Hamad,
  • Raghad Zuhair Yousif,
  • Shahab Wahab Kareem

摘要

The rapid growth of Internet of Things (IoT) ecosystems has created new challenges for building intelligent and privacy-preserving Intrusion Detection Systems (IDS). Traditional centralized solutions are often limited by latency, data imbalance, and exposure of sensitive information. To address these challenges, this study proposes a novel framework called Class-Aware Curriculum Federated Learning (CACFL), which integrates curriculum learning, class-dependent scheduling, and geometric median aggregation within a federated architecture. The proposed framework combines structured, temporal, and contextual features from multimodal IoT data—such as network traffic, device telemetry, and behavioral logs—to enhance intrusion detection performance. It incorporates conditional data augmentation through a CVAE-GAN model and employs an adaptive optimization strategy using the hybrid Bat–Kestrel–Antlion (BKA) optimizer to dynamically tune learning parameters. The framework leverages the complementary strengths of XGBoost, 1D-CNN, and Elastic Weight Consolidation (EWC) models for robust and generalizable feature extraction. Experimental validation demonstrates that CACFL achieves improved detection stability, resilience to non-IID data, and enhanced recognition of minority attack patterns. The study concludes that the proposed approach offers a scalable and privacy-conscious foundation for next-generation intrusion detection in large-scale IoT environments.