Enhancing Phishing Detection A Machine Learning Approach with Feature Selection and Deep Learning Models

Abstract

In the contemporary cybersecurity landscape of 2026, phishing remains a pervasive threat, evolving from simple deceptive emails to sophisticated multi-vector attacks targeting sensitive personal and corporate data. This research presents an Enhanced Phishing Detection Framework that integrates advanced feature selection techniques with highperformance deep learning models to achieve superior classification accuracy. Traditional detection methods often struggle with highdimensional feature spaces, leading to increased computational latency and a higher rate of false positives. Our approach addresses these challenges by implementing a hybrid feature engineering phase, utilizing Permutation Importance and Recursive Feature Elimination (RFE) to identify a compact yet highly discriminative subset of 14–25 key attributes from a dataset of over 58,000 URLs. These features encompass URL lexical structures, HTML content-based patterns, and domain-level security indicators such as SSL certificate age and DNS record consistency.The detection core leverages a suite of deep learning architectures, including Feedforward Neural Networks (FNN), Convolutional Neural Networks (CNN) for structural pattern recognition, and Long ShortTerm Memory (LSTM) networks to capture sequential dependencies in URL strings. By employing a Wide & Deep learning paradigm, the system simultaneously memorizes specific malicious patterns while generalizing to unseen "zero-day" phishing variants. Experimental results demonstrate a remarkable detection accuracy of 99.1% and an F1-score of 98.9%, significantly outperforming baseline machine learning models like SVM and Naive Bayes. Furthermore, the inclusion of Explainable AI (XAI) techniques provides transparency into the model's decision-making process, allowing security analysts to identify specific triggers for flagged URLs. This methodology offers a scalable, real-time solution for securing digital Int. J. Engg. Res.& Sci.& Tech. 2026, ISSN 2319-5991 Vol. 22, No. 2(2), 2026 https://ijerst.org/index.php/ijerst 404 communications, providing a robust defense layer that adapts to the shifting strategies of modern cyber adversaries.