Machine Learning for Malware Analysis, Revisited
Lakhotia, Arun
The classic method of developing classifier models for malware classification has many limitations resulting from challenges in acquiring accurate representative malware data, acquiring representative benign data, and concept drift.
We present a method of creating malware classifiers that addresses these challenges and are more aligned for use in incident response and threat hunting. The method has been implemented in MAGIC (Malware Genomic Correlation) system and has been in use for several years.