期刊名称：International Journal of Engineering and Computer Science
印刷版ISSN：2319-7242
出版年度：2015
卷号：4
期号：3
页码：10992-10997
出版社：IJECS
摘要：Pattern classification systems are commonly used in adversarial applications, likebiometric authentication, network intrusion detection, and spam filtering, in which data can be going onpurpose manipulated by humans to undermine their operation. Extending pattern arrangement[1] theory anddesign methods to adversarial settings is therefore a novel and very relevant research direction, which hasnot yet been pursued in a systematic way. Our address one of the main open issues: evaluating at designphase the security of pattern classifiers, namely, the performance degradation below potential attacks theymay incur during operation. It proposes an algorithm for the generation of training and testing sets to beused for Security evaluation . Developing a framework for the empirical evaluation of classifier securityat design phase that extends the model selection and act evaluation steps of the classical design cycle. Ourproposed framework for empirical evaluation of classifier security that formalizes and generalizes themain thoughts designed in the literature, and give examples of its use in three real applications. report resultsshow that security evaluation can provide a more complete thoughtful of the classifier’s behavior inadversarial environments, and lead to improved design choices .
