Effective Malware Classification using Fine-tuned CNN Architecture: An Image-Based Approach

Malware detection and classification are crucial in cybersecurity to protect systems and networks against malicious attacks. This study introduces a proficient image-based method employing convolutional neural network (CNN) architecture. It aims to enhance malware classification by addressing the challenge of detecting polymorphic or obfuscated malware variants, which can be difficult with conventional techniques reliant on static or dynamic analysis of executable files. Our proposed model tackles this constraint by representing malware samples as grayscale and RGB images, capturing both structural and visual aspects. To improve classification accuracy, we fine-tuned a pre-trained CNN model on a Malign dataset of 25 families and 9342 malware samples, leveraging transfer learning. Our experimental results show that our method accurately classifies distinct types of malware with excellent precision and recall rates. Additionally, image-based representation is resistant to evasion strategies used by advanced malware. Our proposed approach demonstrates potential for enhancing malware detection and classification in practical cybersecurity settings