Malicious website detection under the exploratory attack
Manlin Wang, Fei Hu Zhang, Patrick P. K. Chan · 2013
Malicious websites provide a platform supporting diverse Internet crimes. They do not only steal the sensitive information but also let the hacker to control the computer of users. Malicious website detection with the machine learning technique achieves satisfying result. However, the characteristics of the malicious website may be modified to evade the detection. In this paper, the exploratory attack which misleads the decision of the classifier on the malicious samples by change the feature values with the minimum cost is discussed for malicious website detection. The costs of modifying features of malicious website domain are discussed. The attack model is used to attack the detection system using Support Vector Machine and Fisher Discriminant Classifier. The experimental results show that SVM is more robust than fisher discriminant classifier. Moreover, the vulnerable features are also discussed for the malicious website detection.