Designing and evaluating application layer web threat detection using machine learning techniques

Audience: Undergraduate. -- Dissertation: Thesis (BCS). -- Algoma University, 2017. -- Submitted in partial fulfillment of course requirements for COSC 4235. -- Includes figures. -- Contents: Thesis.

Abstract: This thesis examines the use of machine learning techniques, namely support-vector machines and genetic algorithms, for the purpose of detecting the following application layer web threats: SQL injections, cross-site scripting, and remote le inclusion attacks. Detecting these attacks becomes more important as the Internet grows and leveraging the strengths of machine learning is one of the many potential avenues in order to improve detection. The examination entails using the techniques to detect the aforementioned threats in a collection of unseen web request data and drawing critical conclusions about their strengths, weaknesses, and viability. Through this process several drawbacks to the genetic algorithm approach stood out, more specifically in its error prone detection and high variability of performance; and while the support-vector machine did solve several of these issues and produced great results it's complexity could be troublesome for real-world applications.