Hearthstone AI: Oops to well played

Abstract

Online digital collectible card games have seen a massive rise in popularity recently, none more so than Hearthstone: Heroes of Warcraft. While the game is mainly player vs. player focused, a need for competent game playing AI has arisen as well. This project attempts to tackle this problem by presenting a solution for a game playing AI using a tree-based game simulation and machine learning state evaluator. Additionally, the paper provides a way to breakdown Hearthstone into a feature set usable by machine learning algorithms. The AI based on this solution is shown to perform better than multiple testing solutions when played against them.