Enhanced Malware Detection for IoT Networks Utilizing a 2D-CNN with Data Augmentation on the Malimg Dataset
Iman Fadhil Saleh · Journal of Al-Qadisiyah for Computer Science and Mathematics · 2025
With the swift growth of the Internet of Things (IoT), the attacker’s threat surface has increased multi-fold, making IoT networks a hotbed for multiple malware types. IoT networks’ ever-changing and diverse structures make it almost impossible for traditional malware detection systems to effectively identify and classify hostile traffic in real-time. One viable approach to solving the issue is training Convolutional Neural Networks (CNNs) with data augmentation techniques, as expanding the dataset could help improve IoT malware detection by enabling better classification of distinct malware traffic patterns. In order to enhance detection performance, the authors of this paper suggest a unique two-dimensional (2D) CNN architecture that is tailored for the Malimg dataset and incorporates data augmentation techniques. The suggested method offers a major improvement over other studies that used manually designed dataset-specific characteristics for malware classification by automatically extracting features straight from the infected pictures. By minimizing overfitting, this technique allows the model to train efficiently over a mere ten epochs. The model is better able to generalize and adjust to a greater range of malware samples by employing data augmentation. The outcome of the model built on the Single-CNN architecture performs better than other established models like DenseNet, VGG16, and the VBDN framework obtained an accuracy of 98.86%. The paper included comparative graphs and line plots, bar charts and pie charts demonstrating effectiveness of the proposed model and its original version. These outcomes emphasize the value of domain specific optimizations to solve intricate problems of malware detection and cybersecurity. These results put attention to sophisticated techniques needed to solve malware detection problems in IoT networks and underscore the need for evolving methodologies for these emerging issues.