Phishing and Legitimate Websites Dataset

This dataset provides a comprehensive multimodal benchmark for phishing website detection by combining structured URL-based features with webpage screenshot images. It contains 60,000 website URLs, including 31,641 phishing URLs collected from PhishTank and 28,359 legitimate URLs collected from Tranco. For each URL, 18 lexical and network-based features were extracted using automated Python scripts, capturing characteristics such as URL length, entropy, domain composition, special character usage, digit count, and other structural properties commonly used in phishing detection. To complement the structured data, the dataset also includes 21,311 webpage screenshots captured automatically using a Selenium WebDriver-based browser automation framework. These screenshots represent both phishing and legitimate websites and are organized according to their respective classes. Additionally, the repository includes the publicly available Phish-Iris image dataset containing 2,849 webpage screenshots, enabling comparative evaluation and transfer learning experiments. The dataset supports research in cybersecurity, machine learning, deep learning, computer vision, multimodal learning, explainable artificial intelligence (XAI), and web security. It can be used for phishing website classification, feature engineering, visual representation learning, multimodal fusion, benchmark development, transfer learning, explainability studies, and comparative evaluation of traditional and deep learning methods. The data are provided in standard formats (CSV for URL features and PNG for webpage screenshots), making them readily reusable for academic research, educational purposes, and the development of intelligent phishing detection systems.