Artificial intelligence (AI) in phishing email detection typically focuses on centralized data training that eventually accesses sensitive raw email data from the collected data repository. Due to privacy, organizations are reluctant to share their email data, and it is uncommon to find enough data in one organization to forming a global AI model. Thus, a privacy-friendly AI technique such as federated learning (FL) is a desideratum. FL enables machine learning over multi-organizational email datasets to preserve their privacy without the requirement of accessing them and sharing with other organizations during the learning in a distributed computing framework. To the best of our knowledge, this work is the first to investigate FL in email anti-phishing. Building upon a Recurrent Convolutional Neural Network for phishing email detection, we analyze and evaluate FL-entangled learning performance under various settings, including (i) balanced and imbalanced data distribution among organizations, (ii) scalability, and (iii) communication overhead. Our results positively corroborate comparable performance statistics of FL in phishing email detection to centralized learning. As a trade-off to privacy and distributed learning, FL has a communication overhead of 0.179 GB per global epoch per its organizations, and a gradual degradation in performance if we increase the number of organizations but keep the same total email dataset. However, if we allow to increase the total email dataset with the introduction of new organizations in the FL framework, the organization-level performance is improved. For example, a newly added organization in FL makes an improvement in testing accuracy by 1.87% and fast convergence compared to centralized learning. Besides, our empirical results find that FL has a good performance over imbalanced email dataset.
View on arXiv