Cyber attacks have also increased as smart devices have been used more frequently in recent years. Phishing is a type of fraud in which a person pretends to be someone they can trust by sending emails or using other communication channels to get sensitive information, like login passwords or account information. Here, we contrast AI and profound learning ways to deal with give a framework that is viable at spotting phishing sites through URL investigation, determined to diminish cyber attacks. We show that while tried utilizing URLs from legitimate login pages, existing methodologies had a broad bogus positive rate. Furthermore, via preparing a base model utilizing obsolete datasets and contrasting it with additional ongoing URLs, we exhibit how models lose precision with time utilizing datasets from different years. Phishing Index Login URL (PILU-90K) is a brand-new dataset that consists of 30K phishing URLs and 60K legitimate URLs, including login and index pages. The latest model we propose accomplishes 96.50% exactness on the introduced login URL dataset when Calculated Relapse matched with Term Recurrence - Backwards Record Recurrence (TF-IDF) include extraction.