Evasion attack on text classified training datasets

Abstract

Machine learning algorithms are widespread used in real world training data classification and detection malware. The learning algorithms to detect malware adversarial manipulated training datasets in evasion. The evasion attacker has certain knowledge on training datasets either internal in deploying time attack or external attack do based on adversarial knowledge. Evasion attack targeted document properties features malware. To present this paper, to do an evasion attack on collected text documents using extraction keyword and find mean words using Naive Bayes models . Also to analyses different machine learning algorithms classification on evasion attacked training datasets and discussed defense methods to prevent training dataset from evasion attack.