Advanced Phishing Detection: Leveraging t-SNE Feature Extraction and Machine Learning on a Comprehensive URL Dataset

Phishing attacks continue to pose a major challenge in today's digital world; thus, sophisticated detection techniques are required to address constantly changing tactics. In this paper, we have proposed an innovative method to identify phishing attempts using the extensive PhiUSIIL dataset. The proposed dataset comprises 134,850 legitimate URLs and 100,945 phishing URLs, providing a robust foundation for analysis. We applied the t-SNE technique for feature extraction, condensing the original 51 features into only 2, while preserving high detection accuracy. We evaluated several machine learning algorithms on both full and reduced datasets, including Logistic Regression, Naive Bayes, k-Nearest Neighbors (kNN), Decision Trees, and Random Forest. The Decision Tree algorithm showed the best performance on the original dataset, achieving 99.7% accuracy. Interestingly, the proposed kNN demonstrated remarkable results on feature-extracted data, achieving 99.2% accuracy. We observed significant improvements in Logistic Regression and Random Forest performance when using the feature-extracted dataset. The proposed method offers substantial benefits in terms of computational efficiency. The feature-extracted dataset requires less processing power; thus, it is well-suited for systems with limited resources. These findings pave the way for developing more powerful and flexible phishing detection systems that can identify and neutralize emerging threats in real-time scenarios.