Defending against phishing attacks

Abstract

Phishing and social engineering techniques have evolved alongside technological advancements, posing significant cybersecurity threats by manipulating individuals into revealing sensitive information. Social engineering, a concept dating back to 1894, leverages psychological tactics to deceive users into compromising confidential data, often leading to malware infections or unauthorized access to secure systems. This project investigates the application of machine learning models for phishing URL detection using lexical analysis, aiming to classify URLs as either phishing or legitimate. Key models, including Support Vector Machine (SVM), Random Forest, Gradient Boosting, and K-Nearest Neighbors (KNN), were applied to identify distinct URL features indicative of phishing behavior. Through rigorous experimentation, Gradient Boosting demonstrated the highest accuracy and superior performance metrics, such as precision, recall, and F1-score, outclassing other models. Its effectiveness is largely due to its ability to iteratively enhance accuracy by combining weak learners, making it well-suited for capturing complex patterns within phishing data. This project highlights the potential of Gradient Boosting in robust phishing detection and underscores the need for continued advancements in machine learning to combat the evolving threat of social engineering.