An energy-aware and lightweight spiking neural network framework for real-time intrusion detection in IoT networks
- 1Vellore Institute of Technology
Abstract
The rapid growth of Internet of Things (IoT) network has led to significant security threats due to increased rate of resource use and sophistication of cyberattacks. Traditional deep learning (DL) Intrusion Detection Systems (IDS) utilize large amounts of resources, including energy and, therefore, cannot be implemented on low-powered IoT devices. In this paper, we present an Energy-Aware Lightweight IDS using Spiking Neural Networks (SNN-IDS) to provide high accuracy with respect to intrusion detection while maintaining low energy and memory consumption. This framework uses a hybrid-feature selection mechanism combining Grey Wolf Optimization (GWO) and Mutual Information (MI) to reduce dimensionality and computation costs. In addition, we use SNN for the classification of intrusion events, which allows us to provide low-power, event-driven processing of the input to our system. We evaluate our framework on Bot-IoT, ToN-IoT, and CIC-IoT2023 security datasets, and the results indicate that SNN-IDS outperforms benchmark deep learning models, including CNN-LSTM and Transformer-CNN, in terms of detection performance, energy efficiency, and inference speed. When evaluated on the Bot-IoT dataset, SNN-IDS produces an accuracy of 99.12%, precision of 98.94%, recall of 99.05%, and an F1 score of 98.99%, while on the ToN-IoT dataset it produced 98.76% accuracy and an F1 score of 98.41% and on the CIC-IoT2023 dataset it produced 98.21% accuracy and an F1 score of 97.95%. The SNN-IDS also reduces energy consumption by 42.8% and inference latency by 35.6% when compared to CNN-LSTM based IDS, while maintaining comparable performance in terms of its ability to detect intrusions. Furthermore, the SNN-IDS reduces memory consumption by 38.3%, which makes it a suitable solution for deployment on edge-based IoT devices.
Related
- Parrot dandelion optimizer-based routing and spiking neural network enabled intrusion detection in underwater IoT
- Attention-based intrusion detection using transformer networks on UNSW-NB15
- An EfficientNet hybrid deep learning detection framework of network intrusion with an optimized mixture-of-experts architecture
- Asymmetric class-conditional routing in a federated quantum mixture-of-experts improves imbalanced intrusion detection in edge industrial IoT networks
- HIFN-Mamba: hybrid information flow networks with selective state-space sequence modeling for intrusion detection in the Internet of Medical Things
BibTeX
@article{adhikari2026energyaware,
title = {{An energy-aware and lightweight spiking neural network framework for real-time intrusion detection in IoT networks}},
author = {Priyangshu Adhikari and Logeswari Govindaraj and Tamilarasi Kathirvel Murugan},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-61822-x},
url = {https://doi.org/10.1038/s41598-026-61822-x},
note = {CC BY},
}
RIS
TY - JOUR
TI - An energy-aware and lightweight spiking neural network framework for real-time intrusion detection in IoT networks
AU - Priyangshu Adhikari
AU - Logeswari Govindaraj
AU - Tamilarasi Kathirvel Murugan
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-61822-x
UR - https://doi.org/10.1038/s41598-026-61822-x
AB - The rapid growth of Internet of Things (IoT) network has led to significant security threats due to increased rate of resource use and sophistication of cyberattacks. Traditional deep learning (DL) Intrusion Detection Systems (IDS) utilize large amounts of resources, including energy and, therefore, cannot be implemented on low-powered IoT devices. In this paper, we present an Energy-Aware Lightweight IDS using Spiking Neural Networks (SNN-IDS) to provide high accuracy with respect to intrusion detection while maintaining low energy and memory consumption. This framework uses a hybrid-feature selection mechanism combining Grey Wolf Optimization (GWO) and Mutual Information (MI) to reduce dimensionality and computation costs. In addition, we use SNN for the classification of intrusion events, which allows us to provide low-power, event-driven processing of the input to our system. We evaluate our framework on Bot-IoT, ToN-IoT, and CIC-IoT2023 security datasets, and the results indicate that SNN-IDS outperforms benchmark deep learning models, including CNN-LSTM and Transformer-CNN, in terms of detection performance, energy efficiency, and inference speed. When evaluated on the Bot-IoT dataset, SNN-IDS produces an accuracy of 99.12%, precision of 98.94%, recall of 99.05%, and an F1 score of 98.99%, while on the ToN-IoT dataset it produced 98.76% accuracy and an F1 score of 98.41% and on the CIC-IoT2023 dataset it produced 98.21% accuracy and an F1 score of 97.95%. The SNN-IDS also reduces energy consumption by 42.8% and inference latency by 35.6% when compared to CNN-LSTM based IDS, while maintaining comparable performance in terms of its ability to detect intrusions. Furthermore, the SNN-IDS reduces memory consumption by 38.3%, which makes it a suitable solution for deployment on edge-based IoT devices.
SN - 2045-2322
ER -