Malware Detection Using Efficientnet

Abstract

The quantity, complexity, and variety of malware are all increasing at an alarming rate. Attackers and hackers frequently create systems that can automatically reorder and encrypt their code in order to avoid detection. This paper proposes an improvement in malware detection using a modern neural network model, EfficientNet, determined to achieve higher accuracy and efficiency. The project was implemented using around 2000 samples classified as malicious and benign files imported from the Dike dataset. The portable executable (PE) files were then converted into grayscale images to carry out malware detection using Efficient, an image classification algorithm based on convolutional neural networks. In particular, 4 models - B0 to B3 were implemented in this study. The Agile software development techniques and methodologies were implemented throughout the process.