AI-Driven Phishing Email Detection: A Comparative Review of Ensemble Learning, Explainable AI, and Stylometric Analysis

Phishing email remains a persistent cybersecurity threat, and large language models (LLMs) now allow fluent phishing text tobe produced at scale. This paper is a focused, literature-based comparative review of three recent studies that approach phishing emaildetection from different directions: a soft-voting ensemble with explainable-AI error analysis, a TF-IDF/SVM web platform with LIMEexplanations, and a stylometric classifier evaluated on GPT-4o-generated phishing emails alongside commercial spam filters. The studiesare compared by objective, dataset, feature family, model, evaluation protocol, reported metrics, explainability and stated limitations.Because the studies use different corpora and protocols, their reported figures are presented with attribution and are not ranked. Thecomparison suggests that content/structural detectors and stylometric detectors have been evaluated on largely disjoint data, so whethereither generalizes across conventional and AI-generated phishing is an open question raised by this review, not a verified research gap.The paper proposes a conceptual, unvalidated workflow that combines header/content/URL features, stylometric features, an ensembleclassifier and LIME-based analyst review, and specifies a future experimental protocol with hypotheses, cross-dataset tests and ablations.No new experimental results are reported.