An Ensemble Machine Learning Scheme for Real-time Phishing URL Detection and Browser-Level Deployment

Abstract

Cybercriminals continue exploiting means to perpetrate crimes, using fake links and malicious websites to send invites, display compelling bids, and attempt to gain access to organizations' transaction data to steal or disrupt organizations' operations, resulting in heavy losses. As the digital age advances, cyber threats also advance, with illegitimate web links sent to millions of people, reaching a level of sophistication that results in numerous victims. Thus, a phishing detection scheme was developed; it evaluates Universal Resource Locators (URLs) and classifies them as legitimate (Good) or malicious (Bad) sites. The scheme used an algorithmic process state powered by machine learning (ML) models, demonstrating its effectiveness in accurately recognizing and categorizing URLs. The developed model incorporates essential features to manage vital datasets, used to identify patterns, and provide accurate predictions. Stress testing, load testing, and reaction time analysis were conducted to assess the scheme's scalability and reliability. These tests were essential for determining how well the system performs under various demand scenarios, ensuring that the developed scheme remains responsive and reliable even during peak demand. The scheme handled increased traffic without a decrease in performance. The performance verified that it could manage a range of user input levels while maintaining efficiency. Kaggle datasets were used and implemented in Python for training (30%) and validation (70%). In addition, a password generator and checker were developed to educate organizations about the importance of password combinations and how they are critical to enhancing security. The model informed individuals/organizations on how to efficiently secure data with a superior level of security to protect organizations' business transactions.