Phishing detection using stochastic learning-based weak estimators

Abstract

Phishing attack has been a serious concern to online banking and e-commerce websites. This paper proposes a method to detect and filter phishing emails in dynamic environment by applying a family of weak estimators. Anomaly detection identifies observations that deviate from the normal behavior of a system and is achieved by identifying the phenomena that characterize the “normal” observation. The new observations are classified either a normal or abnormal based on the characteristics of data learnt. Most of the anomaly detection works with the assumption that the underlying distributions of observations are stationary, where this assumption is relevant to many applications. However some detection problem occurs within environments that are non-stationary. One good example to demonstrate the information is by identifying anomalous temperature pattern in meteorology that takes into account the seasonal changes of normal observations. It is necessary that anomalous observations are identified even with the changes or acquire the ability to adapt to the variations in non-stationary environments. Our experimental results show the feasibility and effectiveness of our approach.