Spectrogram based Anomaly Detection Scheme for Internet-of-Things using Deep Convolutional Neural Network
The revolutionary idea of the internet-of-things (IoT) architecture has gained enormous recognition over the last decade, resulting in exponential growth in the networks, connected devices, and the data processed therein. Since IoT devices generate and exchange a massive amount of sensitive data ove...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 語言: | English English |
| 出版: |
2023
|
| 主題: | |
| 在線閱讀: | http://ir.unimas.my/id/eprint/41363/5/Thesis%20Form_Zeeshan%20Ahmad.pdf http://ir.unimas.my/id/eprint/41363/6/Zeeshan%20Ahmad%20ft.pdf |
| 標簽: | 添加標簽 |
| 總結: | The revolutionary idea of the internet-of-things (IoT) architecture has gained enormous recognition over the last decade, resulting in exponential growth in the networks, connected devices, and the data processed therein. Since IoT devices generate and exchange a massive amount of sensitive data over the traditional internet, security has become a prime concern due to the more frequent generation of network anomalies. A network-based anomaly detection system can provide the much-needed efficient security solution to the IoT network by protecting the network entry points through constant network traffic monitoring. Despite enormous efforts by researchers, these detection systems still suffer from lower detection accuracy in detecting anomalies and generate a high false alarm rate and false-negative rate in classifying network traffic. To this end, this thesis proposes an efficient novel Multistage Spectrogram image-based network anomaly detection system using a deep convolution neural network that utilizes a short-time Fourier Transform to transform flow features into spectrogram images. The results demonstrate that the proposed method achieves high detection accuracy of 99.98% with a reduction in the false alarm rate to 0.006% in classifying network traffic. Also, the proposed scheme improves predicting the anomaly instances by 0.75% to 4.82%, comparing the benchmark methodologies to exhibit its efficiency for the IoT network. To minimize the computational and training cost for the model re-training phase, the proposed solution demonstrates that only 40500 network flows from the dataset suffice to achieve a detection accuracy of 99.5%. |
|---|