Affiliations: [a] Department of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Tehran, Iran | [b] Department of Computer Engineering, Islamic Azad University, Tehran, Iran
Correspondence:
[*]
Corresponding author: Hedieh Sajedi, Department of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Tehran, Iran. E-mail:[email protected]
Abstract: Nowadays, World Wide Web is an absolutely necessary and inevitable part
in the life of all people all over the world which they are using Internet
online services routinely to save and manage their sensitive information. On
the other hand, the attackers' tendency is increasing to set up web-based
client-side attacks. Because of the easiness of making these type of
attacks, efficient and accurate detection of malicious web pages is one of
the most important and critical strategies in the network security domain.
In this paper, we present a model for detection of malicious web pages based
on ensemble learning. Our major purpose of applying the idea of ensemble in
the learning process is to provide more learning chance to the data
instances, which are misclassified by previous classifiers. Therefore, it
causes to reduce the classifiers error rate and finally leads to improve the
malicious web pages classification accuracy. For doing this, we propose an
ensemble approach which uses evolutionary reasoning to achieve improvement
of classification accuracy in the detection of malicious web pages. We
assign weight to the base classifiers and use an optimization technique
based on genetic algorithm to select the best committee members of
classifiers to make an optimal ensemble for classification of web pages. The
obtained results from the implementing of our proposed ensemble method with
different base classifiers illustrate that this algorithm leads to the
classification accuracy improvement in all test cases. The most improvement
is related to using SVM as the base classifier, which was 12.5%.
Moreover, the maximum classification accuracy belongs to the proposed
ensemble method using REPTree as the base classifier and it was 95.38%.
