A Malware Detection and Classification using Artificial Neural Networks

A Review

Abstract

The rapid evolution of malware, particularly polymorphic and metamorphic variants, has rendered traditional detection methods, such as signature-based and behavioural detection, increasingly ineffective. This paper's objective is a comprehensive review of Artificial Neural Networks (ANNs) for malware detection and classification via a comprehensive review of the most widely used ANNs. The study focuses on supervised models, unsupervised models, and hybrid architectures across diverse environments. The study results indicate that the supervised models achieve exceptional accuracy (>95%); the unsupervised models offer interpretability and adaptability to evolving threats but face challenges in generalising to unseen data. Conversely, hybrid models combine spatial and temporal feature extraction, achieving 99.4% accuracy, albeit with higher computational costs. This study emphasises the importance of the need for robust frameworks against obfuscation, efficient architectures for resource-constrained environments, and enhanced generalisation across malware families.