What question did this study set out to answer?

The aim is to review AI and ML techniques used for detecting phishing attacks and analyzing cyber threats.

June 13, 2026

Artificial Intelligence-Driven Approaches for Phishing Detection and Cyber Threat Analysis: a Comprehensive Review

Key Points

The aim is to review AI and ML techniques used for detecting phishing attacks and analyzing cyber threats.
Comprehensive analysis of supervised, unsupervised, and deep learning models including CNNs and LSTMs.
Evaluation of classifiers based on accuracy, precision, recall, and F1-score metrics.
Exploration of feature extraction and NLP methodologies for phishing content analysis.
Ensemble learning methods and deep learning architectures achieve detection rates above 97% in controlled environments.
Traditional rule-based detection methods are less effective against evolving cyber threats.
Identified research gaps and limitations of existing models suggest a need for future work on real-time detection systems.

Abstract

The rapid proliferation of digital communication and internet-based services has led to an exponential rise in cyber threats, particularly phishing attacks and social engineering exploits. Traditional rule-based detection mechanisms have proven insufficient in combating sophisticated, evolving threats. This paper presents a comprehensive review of Artificial Intelligence (AI) and Machine Learning (ML) driven approaches employed for phishing URL detection, email classification, and broader cyber threat analysis. We examine supervised, unsupervised, and deep learning models including Decision Trees, Random Forests, Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks, evaluating their effectiveness based on accuracy, precision, recall, and F1-score metrics. The study also explores feature extraction methodologies, publicly available datasets, and the integration of Natural Language Processing (NLP) for semantic analysis of phishing content. Findings indicate that ensemble learning methods and deep learning architectures consistently outperform traditional classifiers, achieving detection rates above 97% in controlled environments. The paper concludes with identified research gaps, limitations of current models, and directions for future work including real-time adaptive detection systems.

AIに質問

Bookmark