Anomaly detection survey for information security
Harsurinder kaur, Husanbir Singh Pannu · 2017
Information security on cloud, mobile devices, social networks and cyber physical systems deal with outlier detection during the data analysis phase. To effectively define the decision boundary for anomaly classification, it is imperative to acquire the data balance among positive class (outliers) and negative class (normal data) of the underlying data distribution. Unfortunately in the real word applications, the data sets are highly unbalanced in nature which results in poor predication performances for the under sampled positive class. In this paper we have performed a comprehensive review on two fundamental ways to handle data imbalance for anomaly detection (a) data centered and (b) algorithmic centered in lieu of different data distributions for information security.