A CUDA-Accelerated Hybrid CNN-DNN Approach for Multi-Class Malware Detection in IoT Networks

The Internet of Things (IoT) ecosystem has attracted widespread attention worldwide because of its profound commercial, economic, and social impact on the daily lives of human beings. Moreover, IoT devices generate large volumes of heterogeneous data, but participating nodes in the network are usually resource-constrained, which makes them attractive targets for cyber threats and attacks. With the rising complexity of modern attacks, it has become essential to develop scalable and efficient solutions that can handle large volumes of data with both speed and accuracy. In this context, GPU-accelerated computing provides a viable solution by offering massive parallel processing power. Using these capabilities, In this work, a CUDA-accelerated hybrid deep learning model is proposed that efficiently detect network attacks in IoT environments. The model incorporates an adaptive preprocessing and feature selection mechanism, distinguishing it from existing approaches. By integrating the strengths of convolutional neural networks for feature extraction and deep neural networks for classification, the proposed framework enhances detection accuracy, scalability, and overall performance. Malware samples are collected from the kitune dataset for experiments. The proposed scheme achieves higher detection and precision rates, that is, 98.41% precision and 98.56% recall. Compared to CPU-only training, the CUDA-enabled model achieved up to a 62% reduction in total training time. These results demonstrate the effectiveness and practical applicability of the CUDA-empowered hybrid model for real-time IoT malware detection.