A Prior-based Transfer Learning Method for the Phishing Detection

In this paper, we introduce a prior-based transfer  learning method for our statistical machine learning  classifier which based on the logistic regression to detect the  phishing sites that relies on our selected features of the  URLs. Because of the mismatched distributions of the  features in different phishing domains, we employ multiple  models for different regions. Since it is impossible for us to  collect enough data from a new region to rebuild the  detection model, we adjust the existing models by the  transfer learning algorithm to solve these problems. The  proposed algorithm was evaluated on a real-world task of  detecting the phishing websites. After a number of  experiments, our proposed transfer learning algorithm  achieves more than 97% accuracy. The result demonstrates  the use of this algorithm in the anti-phishing scenario is  feasible and ready for our large scale detection engine.