Image-Based Malware Detection Using Deep CNN Models

Abstract

Malware or malicious software represents one of the most remarkable threats to cybersecurity, as it compromises the integrity, confidentiality, and availability of computer systems and networks. Traditional malware detection methodologies frequently prove inadequate in identifying innovative and sophisticated malware variants. Deep learning (DL) presents a promising strategy for malware detection by utilizing advanced algorithms that are capable of discerning intricate patterns from extensive datasets. This study presents a model based on deep learning with Convolutional Neural Network (CNN) for malware classification. This research utilized the Malimg dataset, which includes 9,339 malware samples from 25 distinct families. The approach requires resizing malware images to a resolution of 64 x 64 pixels and normalizing these images for model training. The selection of a 64 × 64 size frame reduces network complexity while speeding up training without sacrificing important information. The architecture of the proposed CNN primarily consists of more than one convolutional layer, max-pooling, dropout to mitigate overfitting problem, fully connected layers for achieving better classification results. The proposed model established an impressive accuracy of 96%. For model evaluation, the following measures of accuracy were used: precision, recall, F1-score, and accuracy. This research shows that CNN-based methods can have a high level of effectiveness in detecting obfuscated malware.