Phishing Websites Classification Placed on URL Features and Extreme Machine Learning

Abstract

Phishing attacks have become an increasingly common threat to individuals and organizations alike. The traditional methods used to detect phishing attacks, such as blacklisting known phishing URLs or using heuristics to identify suspicious websites have proven to be limited in their effectiveness. Phishing attackers continuously evolve their tactics, making it difficult for traditional detection methods to keep up. To address this challenge, this study explores the use of machine learning classifiers to uncover illegitimate websites. Specifically, this research utilizes the Multilayer Perceptron and Bernoulli Naive Bayes (NB) classifiers. The feature selection process is performed using a decision tree classifier, which helps to identify the most relevant features for the classification task. To train and test the classifiers, the study collected a dataset of blacklisted and whitelisted websites. Accuracy, precision, recall, and the ROC curve were only few of the measures used to assess the classifier's effectiveness. The results demonstrate the effectiveness of the Multilayer Perceptron and Bernoulli NB classifiers in detecting phishing websites. The feed forward neural network classifier achieved an accuracy of over 82% on the dataset. These results showcase the potential of machine learning techniques in improving the discovering of phishing attacks and reducing further risks of phishing attacks.