UCL Research Paper | Machine Learning for Static Malware Analysis

This report summarizes an NCC Group and UCL Centre for Doctoral Training research project that evaluated multiple machine learning approaches for static classification of Windows PE binaries. Using a dataset of ~74,924 malware and ~32,967 benign samples, the researchers extracted features including PE headers, byte n-grams, control-flow graphs, and API call graphs, and found that a late-fusion ensemble of per-feature models improved detection performance to 98.9% accuracy, demonstrating the promise of multi-modal ensemble methods for malware detection.