To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
IoT Malware Detection Using Hybrid Deep Learning Algorithms |
|
Author |
SUAAD Mohammed Alanzi and Dr. Abdullah J. Alzahrani |
| Citation |
Vol. 24 No. 12 pp. 1-17 |
|
Abstract |
IoT proliferation has caused a revolutionary change in many industries today, through the connection of billions of devices. On the dark side, this exponential growth in the number of devices has brought along significant challenges for cybersecurity. The greatest challenges are driven by sophisticated malware leveraging weak vulnerabilities in IoT devices. Traditional malware detection, which has been largely signature-based, is increasingly inadequate to counter modern, adaptive malware, which is also polymorphic. Accordingly, we develop a new hybrid deep learning model with convolutional neural networks (CNN) for robust feature extraction and recurrent neural networks-long short-term memory (RNN-LSTM) for further extracting sequential dependencies of the network traffic data. Our hybrid approach captures the importance of both spatial features in malicious traffic patterns, as well as temporal dynamics associated with evolving attack behaviors. The architecture utilizes an end-to-end model, with minimum feature engineering on the raw traffic data. Extensive experiments are conducted on NSL-KDD and IoT-23 datasets to test their effectiveness, which are widely regarded benchmarks for intrusion detection and IoT malware analysis, respectively. Compared to classic machine learning and classic deep learning approaches, our hybrid model reached extremely high accuracy of 99.2 percent on NSL-KDD and 99.7 percent on IoT-23. The model further outperformed others in terms of robustness regarding zero-day attack case detection and handling of imbalanced datasets-a typical problem in cybersecurity research. It also focused on the computational efficiency and scalability issues of the model, together with its adaptability to various IoT environments. Results show that there is a good possibility of its application in real-time malware detection in IoT ecosystems with limited resources, such as smart homes, healthcare devices, and industrial IoT. The integration of CNN and RNN-LSTM paradigms in our hybrid model marks a leap toward the mitigation not only of current but also emerging threats against IoT security. This will be followed by extending the model with federated learning for preserving privacy during data analysis and integrating explain ability methods to foster more trust and adoption in operational environments |
|
Keywords |
IoT, Malware Detection, Deep Learning, Hybrid Models, CNN, RNN-LSTM, NSL-KDD, IoT-23, Cybersecurity. |
|
URL |
