Phish-identifier: Machine Learning based classification of Phishing attacks

Phishing attack is one of the most common online attacks that aims to steal important data. It includes sending fake and fraudulent links via emails, text messages, or an instant message. These links are sent from a masquerading attacker posing as a trusted party to the victims. As malicious link is opened at the victim's end, sensitive information can be revealed to the attacker including login credentials. It is an example of highly effective online fraud activity. Correct identification of phishing attack is an important task that can save from revelation of important information of an organization, person, intelligence and/or a government body. This paper shows machine-learning based classification methods based on data collected from 5000 malicious, 5000 legitimate web-pages, and 48 different attributes. The classification is done using Random Tree (RT), Random Forest (RF), Multi-layer Perceptron (MLP), Support Vector Machine (SVM), Radial Based Function (RBF), and K-Nearest Neighbors (KNN) using different percentages of training and testing data for MLP and K-values for KNN. The results are evaluated using precision, recall, f-measure, true-positive, false positive rate, and ROC curves. Results conclude that RF and MLP (with 80% division of train to test data) yields highest accuracy of 98.37% and 96.95%, respectively. However, SVM classified lowest number of (91%) instances correctly.