Performance evaluation of neural network training algorithms in redirection spam detection

Abstract

Redirection spam is a technique whereby a genuine search user is forced to pass through a series of redirections and finally land on a compromised Web site that may present an unwanted content or download malware on his machine. Such malicious redirections are a threat to Web security and must be detected. In this paper, we explore the Artificial Neural Network algorithms for modeling redirection spam detection by conducting the performance evaluation of the three most used training algorithms, namely scaled conjugate gradient (trainscg), Bayesian regularization (trainbr), and Levenberg–Marquardt (trainlm). Our results indicate that the network trained using Bayesian regularization outperformed the other two algorithms. To establish the success of our results, we have used two datasets comprising of 2200 URLs and 2000 URLs, respectively.