Mastering Machine Learning for Penetration Testing

Adversarial features

Feature selection is an important step in every machine learning project. Attackers are also using adversarial feature selection to attack models. I highly recommend that you read the research done by the same team (Pattern recognition and applications lab Italy researchers), presented in a paper called, Is Feature Selection Secure Against Training Data Poisoning?

The team showed that by poisoning the embedded feature selection algorithms, including LASSO, ridge regression, and the ElasticNet, they fooled a PDF malware detector.

There are many Python frameworks and open source projects that were developed by researchers to attack and evaluate machine learning models, like CleverHans, the Adversarial Machine Learning