DESIGN OF INTELLIGENT FEATURE SELECTION TECHNIQUE FOR PHISHING DETECTION

Abstract

Phishing attacks lead to significant threats to individuals and organizations by gaining unauthorized access. The attackers redirect the users to fake websites and steal their credentials and other confidential data. Various techniques are employed to detect phishing using machine learning algorithms or static detection techniques that use blacklisting of web URLs. The attackers tend to change their approach to launch an attack, making it difficult for traditional phishing detection techniques to safeguard the user. The performance of conventional detection methods relies on exhaustive data and features selected for classification. Features selected for designing detection systems majorly contribute to the performance of the detection system. Phishing detection techniques rely mainly on static features that are selected based on traditional feature selection or ranking techniques. This paper proposes an innovative approach to phishing detection by designing a feature selection technique using reinforcement learning. A novel reinforcement learning agent is designed that uses a dynamic, adaptive, and data-driven approach to improve classifier performance in phishing detection. The technique is designed to select the features using the RL agent dynamically. We have evaluated our technique using the real-world phishing dataset and compared its performance with the existing techniques. Based on the evaluation, our proposed methodology of dynamic feature selection gives the best accuracy of 99.07 % with the random forest classifier model. Our work contributes to advancing phishing detection methodology by developing a dynamic feature selection technique.