Seleção de atributos para modelagem preditiva de estratégias em jogos de pôquer

Abstract

Poker is a card game with a very simple objective, basically players bet that their card combination is better than that of their opponents. In the area of ​​computing, more specifically the areas of Artificial Intelligence and machine learning, poker is used as a research object due to its characteristics that make it an interesting challenge for such. It is characterized by a multi-agent competitive environment, partially observable and stochastic. Given this set of characteristics, several machine learning techniques can be applied, in one of the related works the problem of learning the strategies of a player through his records of previous games was addressed. A set of attributes was modeled to represent four aspects of the game: position, aggression, hand quality and current situation. Although some classification techniques had a satisfactory result, the importance of the attributes was not analyzed. Since redundant or irrelevant attributes can impair predictive performance, this work aims to select a subset with the most relevant attributes to be used in the learning process using the attribute selection techniques such as filters and rappers. These techniques aim to reduce the complexity ´ of the model, making it more interpretable, and thus it is expected that the ´ performance of the techniques used will be improved.