Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{181036,
        author = {V.Vijayalakshmi and S Suguna},
        title = {Optimizing Phishing URL Detection with TF-IDF, M-Relief, and RoBERTa: A Deep Learning Approach},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {1},
        pages = {3409-3420},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=181036},
        abstract = {Malicious URLs are a major cyber security 
threat, enabling attacks like phishing and malware. 
Traditional detection methods, such as blacklists and 
heuristics, often miss new or disguised threats. To 
improve detection, machine learning and deep 
learning are increasingly used, though they depend on 
large, regularly updated datasets. This study 
introduces a novel phishing URL classification method 
that combines TF-IDF for feature extraction, Label 
Encoding for transforming categorical data, 
Borderline SMOTE to address class imbalance, M
Relief for feature selection, and RoBERTa, a 
transformer-based deep learning model, for final 
classification. The dataset includes a diverse mix of 
phishing and legitimate URLs. The effectiveness of the 
models is assessed by measuring their accuracy, 
analyzing precision, recall, confidence score, confusion 
matrix, histogram and AUC-ROC specifically for the 
classification of malware attacks. The fine-tuned 
RoBERTa model demonstrates superior performance 
in phishing detection, achieving 98.3% accuracy on the 
test set. Compared to traditional classifiers like 
Random Forest, SVM, and XGBoost, RoBERTa excels 
in identifying phishing URLs with higher precision 
and recall. The proposed approach proves effective for 
real-time phishing detection, enhancing overall cyber 
security protection.},
        keywords = {Borderline SMOTE, TF-IDF, XGBoost,  RoBERTa, Label encoding},
        month = {June},
        }