Development of a Predictive Model for the Detection of CAPTCHA Smuggling Attacks using Supervised Deep Learning based Approach

Abstract

CAPTCHA is a piece of program designed to distinguish human beings from bots. These are computer generated tests which can be solved by humans but will be difficult to be solved by computers. Bots smuggled CAPTCHAs are gradually on the increase in order to deceive unsuspecting users and inadvertently infect systems. From the available literature reviewed so far, there is no model to detect or predict CAPTCHA smuggling attack. The aim of this work is to come up with a model capable of predicting this attack. The approach used was based on deep supervised neural network approach. In order to achieve the aim, framework based on hyperparameter specification was developed. The model was evaluated on the available CAPTCHA smuggling dataset. The accuracy of prediction achieved in this work is 77.89% at consistency of 0.1543. The sensitivity and specificity of the model are 78.11% and 78.2%, respectively