MS-ADS: Multistage Spectrogram image-based AnomalyDetection System for IoT security
The innovative computing idea of Internet-of-Things (IoT) architecture hasgained tremendous popularity over the last decade, resulting in an exponen-tial increase in the connected devices and the data processed in the IoT net-works. Since IoT devices collect a massive amount of sensitive information...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
John Wiley & Sons Ltd
2023
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/42090/1/MS-ADS.pdf http://ir.unimas.my/id/eprint/42090/ https://onlinelibrary.wiley.com/journal/21613915 https://doi.org/10.1002/ett.4810 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The innovative computing idea of Internet-of-Things (IoT) architecture hasgained tremendous popularity over the last decade, resulting in an exponen-tial increase in the connected devices and the data processed in the IoT net-works. Since IoT devices collect a massive amount of sensitive informationexchanged over the traditional internet, security has become a prime concerndue to the more frequent generation of network anomalies. A network-basedanomaly detection system can provide the much-needed efficient security solu-tion to the IoT network by detecting anomalies at the network entry pointsthrough constant traffic monitoring. Despite enormous efforts by researchers,these detection systems still suffer from lower detection accuracy in detect-ing anomalies and generate a high false alarm rate and false-negative rate inclassifying network traffic. To this end, this paper proposes an efficient Multi-stage Spectrogram image-based network Anomaly Detection System (MS-ADS)using a deep convolution neural network that utilizes a short-time FourierTransform to transform flow features into spectrogram images. The resultsdemonstrate that the proposed method achieves high detection accuracy of99.98% with a reduction in the false alarm rate to 0.006% in classifying networktraffic. Also, the proposed scheme improves predicting the anomaly instancesby 0.75% to 4.82%, comparing the benchmark methodologies to exhibit its effi-ciency for the IoT network. To minimize the computational and training costfor the model re-training phase, the proposed solution demonstrates that only40500 network flows from the dataset suffice to achieve a detection accuracyof 99.5% |
|---|