Expert Phishing Detection System
Ayoub Alsarhan, Issa Al-Aiash, Dimah Al-Fraihat, Mohammad Aljaidi, Dena Abu Laila · 2024
Phishing attacks are a threat that continuously emphasizes the need for effective detection methods. This paper presents a machine learning detection method that employs a random forest classifier, which is an adaptable algorithm that can handle a variety of feature sets. The dataset used in this work offers a range of features extracted from various URLs, addressing elements such as URL length, domain structure, and keyword presence. We deployed a computational approach for feature engineering. The random forest classifier is employed for both prediction and to assess feature importance, guiding an automated feature selection process. By doing this, the model's performance is optimized, and the most important features are utilized. The findings of this study show the effectiveness of our method with an impressive accuracy of 98.45% on a test set. The classifier exhibits balanced precision and recall for both phishing and legitimate instances. Moreover, the automated feature selection process improves the interpretation of the model, emphasizing key indicators of phishing attacks. The proposed approach not only produces outstanding outcomes but also offers a flexible and scalable approach to address the dynamically evolving threat landscape of phishing attempts.