Performance Comparison of Machine Learning Models for Phishing Website Detection based on Multilayer Perceptron

Abstract

Phishing is a common cybercriminal activity in which attackers lure people into providing data by posing as genuine websites. Currently, alerts and blacklists are well-used methods of detection though they have been proven less effective in the evolving types of phishing. This paper focuses on the impact of a Multilayer Perceptron (MLP) in minimizing the deficiencies of conventional techniques as a tool of identifying phishing websites. The goal is to improve the identification of new and initially unseen phishing sites by building upon MLP capability to resolve multiple relationships between webpage characteristics including URL, HTML, and HTTP properties. The given experiment is conducted on a total 500 phishing and 500 legitimate websites through different machine learning classifiers such as SVM, k-NN, Decision Trees, Naïve Bayes, and MLP using both 5 and 10-fold cross-validation. The performance of the models is measured using commonly used measures these include accuracy, precision, recall, and F1-score with the MLP having the best performance with 98.1% accuracy. This analysis shows that MLP is the best in improving the detection of phishing threats with the best scalability and adaptability in fighting phishing attacks. Hence, this paper reveals that MLP has a great capacity to enhance real-time phishing detection and minimize false alarm rates to afford a reliable defense against one of the constantly evolving cyber threats.