Esports analytics on PlayerUnknown's Battlegrounds player placement prediction using machine learning

Abstract

PUBG (PlayerUnknown’s Battlegrounds) is a video game that has become popular in the past year. This paper aims to predict the placement of PUBG players during the match by detecting the influential features set that can impact the outcome of the PUBG game and build the best prediction model using a machine learning approach. In this study, the dataset is taken from Kaggle, which has 29 attributes that are categorized into one label (winPlacePerc). The training set has divided into five sets with each set has 6000 instances. The decision tree regression model was applied to find the optimum prediction. Other regression models such as Linear Regression and Support Vector Machine are also utilized to compare with the decision tree model’s result. Based on the result analysis, the walkDistance feature was deemed as the most significant factor influencing the results of a PUBG game. Furthermore, there are other common features obtained from the five datasets that represent the crucial factors which are boosts, DBNOs, killPlace, kills, rideDistance and matchDuration. From the three regression models, the Support Vector Machine model built on the significant features has the best performance in terms of RMSE value while the Decision Tree Regression model has the fastest prediction speed among these regression models.