UrlBERT_DANN model datasets

This dataset contains structured lexical, statistical, and semantic features extracted from real-world phishing and benign URLs. To ensure privacy and reproducibility, all URLs have been irreversibly hashed using SHA256, and no raw URL content is included. The features were generated as part of the <code>UrlBERT_DANN</code> hybrid model pipeline, combining BERT embeddings, FastText representations, and handcrafted URL features.Each row corresponds to a unique URL, with the following components:<code>url_hash</code>: SHA256 hash of the original URLLexical features: length, digit ratio, vowel ratio, subdomain count, etc.Structural features: path depth, query parameters, TLD encodingStatistical metrics: character entropy, special character densityKeyword-based binary indicators: e.g., presence of phishing-related tokens like "login", "verify", etc.<code>label</code>: binary label indicating phishing (1) or benign (0)This dataset supports research on phishing detection, adversarial robustness, and domain-adaptive learning without compromising sensitive data.