A Machine Learning Framework for Strategic Gameplay Optimization in Scrabble Using Reinforcement Learning

Abstract

This paper explores and compare various artificial intelligence-based techniques used for playing classic board game Scrabble. As the world is evolving around finding quick and efficient techniques for the solution of each problem using Machine learning, gaming industry is also a flourishing industry and has grown 9.8% compared to previous year and has generated a revenue of 26.14 billion dollars through global platforms this year which majorly captures youth and many gamers. Most popular games played online includes PUBG, Ludo, cards etc. Scrabble is being a vocabulary enhancing game and less popular game has attracted its intension of research towards it. The implementation of this imperfect information game has been done through Monte Carlo Tree Search algorithm, opponent modelling and Q learning machine learning techniques. Reinforcement learning algorithms are the most suitable algorithm for strategic games. This approach uses Q learning algorithm for the implementation of scrabble game where game theory is the science behind the strategy making for the player playing against human. Experimental results suggest that the Q-learning based agent got higher mean scores and also sharpen decision making efficiency, a bit more than the earlier Scrabble gameplay approaches. Game theory strategy can be categorized into two zero sum and non-zero-sum strategy which could be used for various games and Nash equilibrium.