Published Paper Details:

Phishing attacks continue to be a major concern in the digital world, always changing and taking advantage of both human vulnerability and technological flaws, thereby putting confidential information at risk. To fight against such threats, scientists along with cybersecurity experts have started resorting more often to machine learning (ML) methods which enable automation of detection and mitigation efforts against phishing. Among these methods is Gradient Boosting Classifier which has shown excellent results as it achieved 97.4% accuracy rate and 97.7% F1 score after being trained together with other ML algorithms. This research goes deep into analyzing different attributes of URLs aiming at identifying features that can effectively differentiate legitimate links from malicious ones. These attributes include structural components like URL length, domain age or presence of subdomains as well as lexical elements i.e., text content and obfuscation techniques commonly used during phishing attacks among others. Furthermore, semantic aspects take into consideration contextual understanding of URLs through examination web page contents, SSL certificates plus behavioral patterns associated with bad domains. By utilizing all these diverse characteristics, it becomes possible for machine learning algorithms to finely tune themselves so that they can accurately detect phishing URLs with high levels of precision and recall rates. However, the research also recognizes that there are difficulties and limitations when it comes to identifying phishing URLs with machine learning. These include imbalances in datasets, complex feature engineering, and use of adversarial evasion techniques by advanced attackers. Still, this doesn't mean that Machine Learning is not effective against phishing attacks; in fact, its effectiveness is still significant especially shown by performances of Gradient Boosting Classifier.