Skip to content
Conference

URL-Based Phishing Website Detection Using Machine Learning: A Comparative Study of URL Feature Extraction and XGBoost Optimization

Sep 2026 · 2026 8th International Conference on Cybernetics and Intelligent Systems (ICORIS) · pp. 1-6 · 0 citations · 17 references

Abstract

Phishing remains a persistent cybersecurity threat because malicious URLs can be changed rapidly to evade blacklist- and rule-based defenses. This paper evaluates a lightweight phishing detection pipeline that operates only on information derived from the URL string. Experiments use the PhiUSIIL dataset with 235,795 records (134,850 phishing and 100,945 legitimate). After cleaning, 18 lexical, structural, and security-related URL features were extracted automatically and ranked using tree-based feature importance; a predefined top15 subset was retained for the comparative experiment. Decision Tree, Random Forest, Linear Support Vector Machine, and XGBoost were evaluated on the same 80:20 holdout split. The initial XGBoost model achieved the best test performance: 99.508% accuracy (reported as 99.51%), 99.285% precision, 99.859% recall, 99.571% F1-score, and a 0.961% false positive rate. Grid-search optimization produced a best cross-validation score of 0.995383, but its holdout accuracy (99.493%) was slightly lower than that of the initial model; therefore, the initial XGBoost configuration was retained for deployment. The resulting model was integrated into a URL-input prototype using the same saved feature definitions and feature order. The results support URL-only machine learning as a fast first-level screening mechanism, while also showing that dataset-specific feature importance and single-dataset validation limit claims of generalization.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.