Estudo do jogo de pôquer utilizando métodos estatísticos e aprendizado de máquina

Abstract

The present work was developed with the purpose of modeling actions of players in the game of poker and classify them as winners or losers using statistical and machine learning methods. To apply these techniques, we used a real database that contains around 22 million games played at online casinos. We appeal to Holdem Manager software for data pre-processing and application in algorithms was performed in R, considering 13 explanatory variables that represent players actions. We create a data set in which the ratio of winning and losing players is approximate the same, then the base was divided into 80% for training and 20% for testing and validate the model. The final classification model using Random Forest obtained an accuracy ratio of 0.75. We identified that strategic actions are important for winning in poker, that is, players who tend to be winners behave differently from losers with regard to the frequency of making moves in the game.