Phishing is one of the most common threats on the Internet. Traditionally, detection methods have relied on blacklists and heuristic rules, but these approaches are showing their limitations in the face of rapidly evo...Phishing is one of the most common threats on the Internet. Traditionally, detection methods have relied on blacklists and heuristic rules, but these approaches are showing their limitations in the face of rapidly evolving attack techniques. Artificial Intelligence (AI) offers promising solutions for improving phishing detection, prediction and prevention. In our study, we analyzed three supervised machine learning classifiers and one deep learning classifier for detecting and predicting phishing websites: Naive Bayes, Decision Tree, Gradient Boosting and Multi-Layer Perceptron. The results showed that the Gradient Boosting Classifier performed best, with a precision of 96.2%, a F1-score of 96.6%, recall and precision of 99.9% in all classes, and a mean absolute error (MAE) of just 0.002. Closely followed by the Gradient Boosting Classifier with a precision of 96.2% and a score of 96.6%. In contrast, Naive Bayes and the Decision Tree showed a lower accuracy rate. These results underline the importance of high accuracy in these models to reduce the risk associated with malicious attachments and reinforce security measures in this area of research.展开更多
文摘Phishing is one of the most common threats on the Internet. Traditionally, detection methods have relied on blacklists and heuristic rules, but these approaches are showing their limitations in the face of rapidly evolving attack techniques. Artificial Intelligence (AI) offers promising solutions for improving phishing detection, prediction and prevention. In our study, we analyzed three supervised machine learning classifiers and one deep learning classifier for detecting and predicting phishing websites: Naive Bayes, Decision Tree, Gradient Boosting and Multi-Layer Perceptron. The results showed that the Gradient Boosting Classifier performed best, with a precision of 96.2%, a F1-score of 96.6%, recall and precision of 99.9% in all classes, and a mean absolute error (MAE) of just 0.002. Closely followed by the Gradient Boosting Classifier with a precision of 96.2% and a score of 96.6%. In contrast, Naive Bayes and the Decision Tree showed a lower accuracy rate. These results underline the importance of high accuracy in these models to reduce the risk associated with malicious attachments and reinforce security measures in this area of research.
