Optimization of rulesets with reinforcement learning

Sammanfattning: Security is becoming more and more important for businesses, governments and individuals. Because of the trend of connecting everything to the internet and saving more and more information which can affect both the user and the business if it were released. An unintentional release, a security breach can result in serious economical and reputational losses for all the affected parties. For example, during 2018 in the United States the estimated combined loss from the reported security breaches reached 2.7 Billion. One of the most common ways to protect your information against adversaries is with firewalls. How well a firewall can protect against adversaries depends on how the ruleset of the firewall is configured and the literature shows that configuring a firewall without errors is not an easy task.  Therefore, a closer look is taken on the optimization of firewalls rulesets with a usability metrics as criteria. To optimize the ruleset for a better usability machine learning will be used. The optimization will focus on one aspect of usability within the metric, the perceived human complexity. Around the usability metric a proof of concept application is coded and tested against five real rulesets to see how much the rulesets usability can be improved using machine learning.