Abstract:
"
Operating in dynamic real world spam domain requires more robust adversarial-aware
designs as it is identified as never ending game between learner and attacker. While
most of the researchers are focused on either evasion or poison resistant, there are only
few researches which are focused on both test and train time attacks. In this research,
Ensemble approach has been applied to combine multiple learners which were secured
against single attack based on attack influence. It is compared with individual learners
and results have shown an attack which is targeted on single learner does not imply the
same on Ensemble methods. The proposed method provides motivation for future
works of area in ensemble learning under spam detection."
