Performance of Deep Learning models for phishing detection in controlled environments: A systematic review

Abstract

Phishing remains one of the most persistent threats in the field of cybersecurity, driving the development of Deep Learning (DL) models to enhance their detection. This Systematic Literature Review aims to identify, categorize, and analyze DL models applied to phishing attack detection in controlled environments. Based on the analysis of recent studies, these models have demonstrated strong performance, particularly in metrics such as Accuracy and F1- score. The review also examines commonly used architectures such as RNNs, CNNs, and hybrid models, the types of attack vectors addressed, and the experimental conditions under which the models were evaluated. However, recurring limitations were identified, including the use of non-representative datasets, a lack of standardization in evaluation metrics and attack vectors, and limited validation in real-world scenarios. This review offers a structured synthesis of the current state of DL-based phishing detection and serves as a reference point for future research aimed at improving model performance and practical applicability.