DETECTION OF PHISHING LINK ON WEB APPLICATIONS USING ARTIFICIAL INTELLIGENCE (AI)

Abstract This project presents the design of an AI-powered system for detecting phishing links, addressing the growing threat of sophisticated cyberattacks within Nigeria’s expanding digital economy. Traditional blacklist-based defenses are increasingly ineffective, as attackers rapidly generate new domains to evade detection. To counter this challenge, a dataset was compiled from reputable sources including PhishTank, OpenPhish, URLhaus, and legitimate Alexa Top sites. Lexical and host-based features were extracted from URLs, forming the basis for machine learning analysis. Multiple models, Decision Tree, Random Forest, and Support Vector Machine (SVM) were implemented and evaluated, following the methodology outlined in Sahingoz et al. (2019). Among these, the Random Forest classifier achieved superior performance, recording approximately 95% accuracy with high precision and recall. This minimized false positives while ensuring reliable identification of phishing attempts (Verizon, 2024). The trained model was integrated into a functional web application prototype that provides users with instant safety assessments of submitted links. Results confirm that an AI-driven, feature-based approach is more adaptive and effective against evolving phishing strategies compared to static blacklist methods. This research recommends the adoption of the solution by financial institutions, e-commerce platforms, and educational organizations in Nigeria. Such deployment would enhance cybersecurity resilience, strengthen user trust, and mitigate risks related to financial fraud and identity theft.