Scaling Multi-Objective Optimization for Clustering Malware

Malware continues to be a growing concern in cyber-security. The evolving landscape of obfuscation techniques used by malware authors causes major problems in the identification, analysis and modelling of malware by security analysts. Clustering malware variants based on their behavior has potential to create better malware analysis systems. In this work, we cluster samples from the EMBER dataset that contains features extracted from malevolent Windows portable executable files. We expand on well-known algorithms through extensive testing and discuss the results of the top performing algorithms in terms of clustering metrics and Weighted F1-score.