Identifying Malicious VoIP Usage using Computational Intelligence

Abstract

VoIP user accounts are a prime target for hackers to compromise for profit. VoIP accounts are targets of the same types of attacks as any other Internet account that is authorized with a username and password. Unlike many other Internet accounts VoIP has a direct

monetary cost to the user being compromised. Toll-fraud perpetrated using a compro- mised VoIP account can accrue expensive toll-charges that either the user or the service

provider are liable to pay for. This paper discusses the prior research in detecting unau- thorized usage on VoIP accounts. The researched methods are based on machine learning

techniques. A new technique of using a Recurrent Neural Network for detecting unau- thorized usage periods on a VoIP account is developed and demonstrated. The technique

uses a Long-Short Term Memory style of Recurrent Neural Network to achieve over a 99% accuracy when testing against calls tagged as occurring during a toll-fraud event.