Enhanced Malware Classification: A Hybrid Model Utilizing Denoising Autoencoder and CNN based on visualization method

Thippireddy Harika^1,*, Gera Pradeepini²

¹M. Tech, Dept of CSE, Koneru Lakshmaiah Educational Foundation, Vaddeswaram, Guntur, A.P, India

²Professor, Dept CSE, Koneru Lakshmaiah Educational Foundation, Vaddeswaram, Guntur A.P, India

Emails: thippireddyharika@gmail.com; pradeepini_cse@kluniversity.in

Abstract

In the last few years, technology has developed so rapidly that many malware applications are available in the software market. Cybercrimes are increasing day by day with the usage of malware applications. Traditional approaches are not as effective in detecting malware. This study introduces a novel method for distinguishing malware from benign software applications using deep learning models like Denoising Autoencoder and Convolutional Neural Network. Initially, we extract binary code from the applications and transform it into grayscale images. Then, utilizing a denoising autoencoder, we improve the quality of the grayscale images by eliminating noise, and the Convolutional Neural Network uses processed images as input. Finally, the Convolutional Neural Network is employed to differentiate between malicious and benign applications. We test this methodology on the dataset that contains 10,810 malware and 1082 benign files. The suggested model obtains an accuracy of 97% and an F1-score of 96% and performs better than some traditional methods.

Keywords: Cybersecurity; Radare2; Denoising Autoencoder; Convolutional Neural Network; Malware Classification