Hybrid CNN-MLP for Malware Detection on Heterogeneous Features

Didi, Sabre and Nitschke, Geoff (2025) Hybrid CNN-MLP for Malware Detection on Heterogeneous Features, Proceedings of 3rd EAI International Conference on Safe, Secure, Ethical, Responsible Technologies and Emerging Applications, 8-10 December, Cape Town, South Africa, Springer.

Text 2025-hybrid-cnn-mlp-for-malware-detection-on-heterogeneous-features.pdf - Published Version Download (559kB)

Abstract

Malware attacks remain a critical cyber-security concern, necessitating robust detection solutions for both individuals and organizations. Deep learning methods, including Convolutional Neural Networks (CNNs) and Multi-Layer Perceptrons (MLPs), have become integral to malware detection and classification. However, real-world malware datasets, such as Elastic Malware Benchmark for Empowering Researchers (EMBER), present heterogeneous features that differ in type, scale, and representation. This heterogeneity poses a significant challenge for traditional deep-learning methods, which often assume homogeneity in the input features, resulting in suboptimal learning and detection performance. Moreover, the rapid evolution of malware strains introduces further complexity, as conventional models struggle to adapt to novel patterns. This study proposes a hybrid Multi-Branch MLP and CNN model that handles heterogeneous malware feature groups through dedicated branches to better capture local and global patterns. The model adapts to evolving, heterogeneous malware data and outperforms conventional models, as demonstrated using the EMBER dataset.

Item Type:	Conference paper
Subjects:	Computing methodologies > Machine learning > Learning paradigms > Supervised learning > Supervised learning by classification Applied computing
Date Deposited:	15 Oct 2026 06:58
Last Modified:	15 Oct 2026 06:58
URI:	https://pubs.cs.uct.ac.za/id/eprint/1817

Actions (login required)

View Item