MOJI: Character-level convolutional neural networks for Malicious Obfuscated JavaScript Inspection

Minato Ishida, Naoshi Kaneko, Kazuhiko Sumi

Abstract: Highlights•Character-level convolutional neural networks are robust to obfuscated JavaScript.•A simple convolutional neural network outperforms abstract syntax tree-based methods.•Surprisingly, antivirus software tends to be fooled by additional obfuscation.