Data Exploration

[gregdhill]

In the paper, a distinct subset of cryptographic algorithms & opcodes were correlated which resulted in our model achieving 96% accuracy. It would be advantageous to explore different settings which could further improve accuracy / generalization:

• Trial different feature engineering techniques.
• Build more datasets with a different number of embeddings and/or alternate opcodes.
• Add new cryptographic signatures into the sample pool (following our format specification).