Imbalanced Learning in Massive Phishing Datasets

Phishing is one of the major threats facing internet users in today’s work. Such attacks continue costing billions of dollars to companies around the words thus requiring more efficient detection techniques to curb the danger. This paper proposes a big data friendly implementation of Multiclass Imbalance Learning in Ensembles through Selective Sampling (MILES) that detects phishing attacks with high accuracy. The proposed method is compatible with SPARK, can be trained on a cluster of nodes in parallel, thus reduce the training time by increasing the size of the cluster. In addition, a comparative study of classic machine learning techniques like Random Forest, Naive Bayes, and Decision Trees show that the proposed MILES method provides significantly higher precision and recall.