A Novel Neural-Based Malware Detection Approach for Embedded Devices

In recent years, deep learning-based malware detection systems have been widely used in the field of cybersecurity, especially to detect malicious files in critical infrastructures. This paper proposes a lightweight malware detection method that can identify recent malware while being suitable for execution on embedded devices. This method combines the features of convolutional neural networks (CNN) with the tokenization advantages of transformer models. The proposed model architecture combines a diverse set of neural network layers designed to capture intricate patterns indicative of malicious behavior in files. Extensive experiments conducted using the latest CIC- Malmem-2022 dataset demonstrate that our method surpasses existing machine learning-based models in the literature for both malware detection and specific attack type identification. The model achieved a test accuracy of 99.97, demonstrating its high effectiveness in distinguishing between benign and malicious files. By combining dense neural networks, specialized capsule layers, and attention mechanisms, our architecture offers a robust solution for malware detection.