Annotation: The problem of protecting machine learning models used in intrusion detection systems from adversarial attacks is considered. Possible methods of protection against adversarial samples based on data anomaly detectors and an autoencoder are analyzed. The results of an experimental study of protective mechanisms that demonstrated high efficiency in detecting distorting data using a Random Forest model are presented.

Keywords: adversarial attack, machine learning security, adversarial sample detection, machine learning, intrusion detection system, Random Forest