Predicting and Enhancing Hearthstone Strategy with Combinatorial Fusion
General Material Designation
[Thesis]
First Statement of Responsibility
Gorelick, Henry William
Subsequent Statement of Responsibility
Hsu, D. F.
.PUBLICATION, DISTRIBUTION, ETC
Name of Publisher, Distributor, etc.
Fordham University
Date of Publication, Distribution, etc.
2020
GENERAL NOTES
Text of Note
62 p.
DISSERTATION (THESIS) NOTE
Dissertation or thesis details and type of degree
M.S.
Body granting the degree
Fordham University
Text preceding or following the note
2020
SUMMARY OR ABSTRACT
Text of Note
The goal of this master's thesis is to demonstrate that combinatorial fusion analysis (CFA) can effectively predict winners and enhance play strategy of Blizzard Entertainment's collectible card game Hearthstone. CFA is used to combine and evaluate the performance of the combinatorial combinations of five machine learning models trained on 500 Hearthstone game simulations. For each combinatorial combination, the score function of the score combination and the score function of the rank combination is derived for each of the five models, and the performance of each is compared and evaluated. The improvement in performance of certain combinations over the individual components validates that CFA is an effective method for predicting the winner of Hearthstone games and enhancing play strategy. Furthermore, the resulting models could be used to boost Monte Carlo Tree Search and implement a competitive Hearthstone playing AI agent.