Spam Detection Based on Feature Evolution to Deal with Concept Drift

Electronic messages are still considered the most significant tools in business and personal applications due to their low cost easy access. However, e-mails have become a major problem owing high amount of junk mail, named spam, which fill e-mail boxes users. Several approaches been proposed detect such as filters implemented servers user-based spam message classification mechanisms. A with these is detection presence concept drift, especially result changes features over time. To overcome this problem, work proposes new system based on analyzing evolution features. The method divided into three steps: 1) model training; 2) drift detection; 3) knowledge transfer learning. first step generates models, commonly conducted machine second introduces strategy avoid drift: SFS (Similarity-based Features Se- lection) that analyzes taking account similarity obtained between feature vectors extracted from training data test data. Finally, third focuses following questions: what, how, when acquired knowledge? evaluated using two public datasets. results experiments show it possible infer threshold (drift) order ensure updated through transfer. Moreover, our anomaly able perform parallel independent tasks.