Abstract:
In the current year, the cyber attackers use many kinds of phishing attacks, such
as fake websites or links, emails, and messages, to obtain private information. The
phishing detection system is very important to control cybercrimes. The proposed
system applies the ANOVA feature selection method based on the UCI dataset. The
ANOVA selects the most relevant twenty-six features from thirty features of the
dataset. The system combines three machine learning classifiers: Random Forest,
Gradient Boosting, and Support Vector Machine for improved performance using two
ensemble learning methods: hard voting and soft voting. And then three classifiers are
supported to produce a more accurate Phishing URL Detection System. The final
predictions use the ensemble learning techniques to classify URLs as phishing or
legitimate. In this study, the testing accuracy of the UCI dataset is obtained at 96.88%
for soft voting and 96.65% for hard voting, displaying the performance of the proposed
system. Additionally, the open URL testing accuracy is achieved at 94.07% for soft
voting and 94% for hard voting. It is implemented by the Python programming
language to support protecting users from phishing threats. The proposed model would
be practical for helping cybersecurity specialists and the public can accurately detect
phishing attacks.