A Privacy-Preserving Machine Learning Ensemble for Spam Detection

In this era of spam and fraudulent emails, it is important to detect spam to reduce the chances of falling victim to fraud. Many email applications such as Gmail, and Outlook run a spam detection algorithm to direct spam mail to the spam folder. However, tricksters learn to escape the spam box by studying the behavior of these standard algorithms. This research study presents a module that can be used to improve the performance of spam filters. This module can be integrated as an extension to these spam filters. The module accepts only encrypted emails to eliminate privacy concerns. Since no copies of the decrypted emails are maintained on the server, the emails are not available to the spam filter module. This feature is deemed important as privacy concerns are getting louder. The spam detector uses multiple machine-learning algorithms including a bagging classifier. It also features a voting classifier to finalize the classification of mail into ham or spam. The ensemble is tested with two datasets and a procedure to build a framework for constructing best-performing models has been discussed.