International Research Journal of Innovations in Engineering and Technology - IRJIET International Open Access, Monthly, Peer Reviewed, Reputed Journal ISSN (online): 2581-3048

Impact Factor (2025): 6.9

DOI Prefix: 10.47001/IRJIET

EdgeMind: A Lightweight Federated Learning Framework for Real-Time IoT Anomaly Detection on Resource-Constrained Edge Devices

Swayamdip Devendra ChaurpagarStudent, B.Tech - Computer Science and Engineering, Shri Sai College of Engineering and Technology, DBATU University, Bhadrawati, Chandrapur, Maharashtra, IndiaYash Purushottam BokdeStudent, B.Tech - Computer Science and Engineering, Shri Sai College of Engineering and Technology, DBATU University, Bhadrawati, Chandrapur, Maharashtra, IndiaSomnath Anna KadamStudent, B.Tech - Computer Science and Engineering, Shri Sai College of Engineering and Technology, DBATU University, Bhadrawati, Chandrapur, Maharashtra, IndiaHarsh Vinod KolarkarStudent, B.Tech - Computer Science and Engineering, Shri Sai College of Engineering and Technology, DBATU University, Bhadrawati, Chandrapur, Maharashtra, IndiaSuraj S. BankarAssistant Professor, Computer Science and Engineering, Shri Sai College of Engineering and Technology, DBATU University, Bhadrawati, Chandrapur, Maharashtra, India

Vol 10 No 5 (2026): Volume 10, Issue 5, May 2026 | Pages: 160-169

International Research Journal of Innovations in Engineering and Technology

OPEN ACCESS | Research Article | Published Date: 09-05-2026

doi Logo doi.org/10.47001/IRJIET/2026.105021

pdf iconFull Text PDF
Abstract

The unprecedented proliferation of Internet of Things (IoT) deployments — from industrial control systems and smart grids to healthcare wearables and autonomous vehicle networks — has generated an insatiable demand for real-time anomaly detection at the network edge. Transmitting raw sensor telemetry to centralised cloud servers for inference introduces unacceptable latency (120-850 ms round-trip over LTE), creates single points of failure, and violates increasingly stringent data-sovereignty regulations including India's Digital Personal Data Protection Act 2023 and the EU General Data Protection Regulation. FederaFederated Learning (FL) offers a compelling alternative: models are trained locally on edge devices and only gradient updates — never raw data — are shared with an aggregation server. However, existing FL frameworks (Flower, PySyft, TensorFlow Federated) are engineered for data-centre-class hardware and impose memory footprints (512 MB+) and communication overheads incompatible with the constrained microcontrollers (256 KB-2 MB RAM) that constitute the majority of deployed IoT edge nodes. This paper presents EdgeMind, a novel lightweight federated learning framework purpose-built for resource-constrained IoT edge devices. EdgeMind introduces three key contributions: (1) a gradient compression scheme — Sparse Top-k Gradient Pruning with Adaptive Threshold (STGPAT) — that reduces per-round communication volume by 94.3% while retaining 97.1% of convergence accuracy; (2) a heterogeneity-aware asynchronous aggregation protocol (HAAP) that tolerates device dropout rates up to 40% without accuracy degradation; and (3) an on-device anomaly detection model (EdgeMind-Net) — a quantised MobileNetV3-inspired 1D-CNN architecture occupying 187 KB RAM — deployable on Raspberry Pi Zero 2W, ESP32-S3, and Arduino Nano 33 BLE Sense. Evaluation on three benchmark datasets (UNSW-NB15, KDD Cup 99, and WADI) demonstrates F1-scores of 0.943, 0.961, and 0.927 respectively, with end-to-end inference latency of 4.2 ms on Raspberry Pi Zero 2W — a 28x improvement over the nearest FL baseline framework. EdgeMind is open-source and available at github.com/edgemind/edgemind-framework.

Keywords

Federated Learning, IoT Anomaly Detection, Edge Computing, Gradient Compression, Resource-Constrained Devices, Privacy-Preserving Machine Learning, Quantised Neural Networks, 1D-CNN, Heterogeneous Federated Learning, Smart Grid Security, DPDP Act 2023.


Citation of this Article

Swayamdip Devendra Chaurpagar, Yash Purushottam Bokde, Somnath Anna Kadam, Harsh Vinod Kolarkar, & Suraj S. Bankar. (2026). EdgeMind: A Lightweight Federated Learning Framework for Real-Time IoT Anomaly Detection on Resource-Constrained Edge Devices. International Research Journal of Innovations in Engineering and Technology - IRJIET, 10(5), 160-169. Article DOI https://doi.org/10.47001/IRJIET/2026.105021

Licence Copyright (c) 2026 International Research Journal of Innovations in Engineering and Technology creative common licence This work is licensed under Creative common Attribution Non Commercial 4.0 Internation Licence
References
  1. IoT Analytics Research, 'State of IoT 2024: Number of Connected IoT Devices Growing 13% to 18.8 Billion Globally,' IoT Analytics GmbH, Hamburg, Germany, May 2024.
  2. Claroty, 'Global State of Industrial Cybersecurity 2023: Critical Infrastructure at Risk,' Claroty Ltd., New York, USA, 2023.
  3. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, 'Communication-Efficient Learning of Deep Networks from Decentralized Data,' in Proc. 20th AISTATS, Fort Lauderdale, FL, 2017, pp. 1273-1282.
  4. T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, 'Federated Optimization in Heterogeneous Networks,' in Proc. MLSys 2020.
  5. J. Konecny, H. B. McMahan, F. X. Yu, P. Richtarik, A. T. Suresh, and D. Bacon, 'Federated Learning: Strategies for Improving Communication Efficiency,' arXiv:1610.05492, 2016.
  6. Y. Lin, S. Han, H. Mao, Y. Wang, and W. J. Dally, 'Deep Gradient Compression: Reducing the Communication Bandwidth for Distributed Training,' in Proc. ICLR 2018.
  7. A.Khraisat, I. Gondal, P. Vamplew, and J. Kamruzzaman, 'Survey of intrusion detection systems: techniques, datasets and challenges,' Cybersecurity, vol. 2, no. 1, Art. 20, 2019.
  8. R. C. Staudemeyer and E. R. Morris, 'Understanding LSTM — a tutorial into Long Short-Term Memory Recurrent Neural Networks,' arXiv:1909.09586, 2019.
  9. Z. Wang, W. Yan, and T. Oates, '1D Convolutional Neural Networks and Applications: A Survey,' Mech. Syst. Signal Process., vol. 158, Art. 107832, 2021.
  10. A.Howard et al., 'Searching for MobileNetV3,' in Proc. IEEE/CVF ICCV, Seoul, 2019, pp. 1314-1324.
  11. T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, 'FedProx: Federated optimization for heterogeneous networks,' in Proc. MLSys 2020.
  12. S. Caldas et al., 'LEAF: A Benchmark for Federated Settings,' arXiv:1812.01097, 2018.
  13. C. Briggs, Z. Fan, and P. Andras, 'Federated learning with hierarchical clustering of local updates to improve training on non-IID data,' in Proc. IJCNN, 2020, pp. 1-9.
  14. C. He, A. D. Shah, Z. Tang, D. F. A. N. Sivashunmugam, K. Bhowalkar, A. Ryou, G. Li, L. Shen, and M. Annavaram, 'FedML: A Research Library and Benchmark for Federated Machine Learning,' arXiv:2007.13518, 2020.
  15. K. Bonawitz et al., 'Practical secure aggregation for privacy-preserving machine learning,' in Proc. ACM CCS, 2017, pp. 1175-1191.
  16. B. Jacob et al., 'Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference,' in Proc. IEEE CVPR, Salt Lake City, 2018, pp. 2704-2713.
  17. N. Moustafa and J. Slay, 'UNSW-NB15: a comprehensive data set for network intrusion detection systems,' in Proc. MilCIS, Canberra, 2015, pp. 1-6.
  18. M. Tavallaee, E. Bagheri, W. Lu, and A. A. Ghorbani, 'A detailed analysis of the KDD CUP 99 data set,' in Proc. IEEE CISDA, 2009, pp. 1-6.
  19. J. Goh et al., 'A Dataset to Support Research in the Design of Secure Water Treatment Systems,' in Proc. CRITIS, Springer, 2016, pp. 88-99.
  20. P. Blanchard, E. M. El Mhamdi, R. Guerraoui, and J. Stainer, 'Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent,' in Proc. NIPS, 2017, pp. 119-129.

Copyright @ 2026 IRJIET. All Right Reserved.

Licensed underCreative Commons Attribution 4.0 International License.