Evaluating the Efficiency of Fine-Tuned Machine Learning Models in Phishing URL Detection

Abstract

Phishing URL detection is a critical challenge in cybersecurity, particularly in the context of social networks, which are widely used for information dissemination. Cybercriminals are increasingly exploiting these platforms to launch phishing attacks to deceive users into revealing sensitive information. Researchers employed various machine learning techniques to detect phishing URLs. However, phishing tactics continue to evolve, requiring more accurate and robust detection systems. This study addresses the challenge by evaluating the effectiveness of various machine learning models in detecting phishing URLs. The dataset, sourced from Kaggle, contains 11,054 instances with 32 features representing URL attributes. Models such as Logistic Regression, k-Nearest Neighbors, Naive Bayes, Decision Trees, and Random Forest were assessed for their performance. Hyperparameter tuning was applied to optimize the models, resulting in significant accuracy, precision, recall, and F1 score improvements. Among these models, the Random Forest model, with optimal hyperparameters, achieved the highest accuracy of 97.26%, outperforming other models. This research highlights the importance of hyperparameter tuning in enhancing machine learning models for phishing URL detection and underscores its potential to strengthen cybersecurity measures against phishing threats.