An Effective Phishing Detection Model using Hybrid Features Algorithm and Deep Learning Techniques

Abstract

Phishing detection is one of the important and demanding tasks in recent newlinedays for avoiding the illegal activities and information theft. Typically, the newlinephishing is a kind of squeezing, where the attacker is termed as phisher who newlineintends to misuse the authorized confidential data of the web users. The general newlineinformation whipped by the phishing attacker are the user name and password newlinedetails of the individual user, account number, e-banking information, and credit newlinecard details. So, it is highly important to prevent the user information and system newlineagainst the harmful phishing attacks. For this purpose, different types of newlinemechanisms have been developed in the conventional works to detect the newlinephishing attacks. The major drawbacks of the existing techniques are high time newlinecomplexity, and reduced accuracy.In order to solve these problems, enhanced security techniques are newlinedeveloped in the proposed work, which helps to detect the website phishing newlineattacks with increased detection accuracy. An enhanced Bidirectional long short- newlineterm memory (BiLSTM) Repetitive Neural Network (RNN) is developed in the newlinefirst module of this work for identifying the web phishing attacks. Then, an newlineeXtreme Gradient Boost (XGB) classification technique is utilized for detecting newlinethe phishing attacks based on the set of multiple features.Moreover, the Dynamic Threshold Algorithm (DTA) is deployed for newlinecategorizing the types of phishing attacks by incorporating the hybrid and visual newlinesimilarity features. Moreover, the performance of these techniques are validated newlineand compared by using different evaluation metrics. The results are compared with the recent state-of-the-art methods for proving the betterment of the proposed technique.