การทำนายผลชนะก่อนการเล่นเกมโดยใช้กระบวนเรียนรู้แบบอองซอมเบลอในเกมอาร์โอวี

Abstract

Realm of Valor (RoV) is a famous MOBA (Multiplayer Online Battle Arena) game with an average of 25 million players per day in Thailand alone. In addition, the game also organizes international events with a total prize pool worth millions of US dollars. However, the game is very complicated. Players must be highly experienced to win, especially the hero selection process that each player must do in the early game. Because the hero chosen can affect the outcome of the game, but there are still many heroes to choose from. This study compares machine learning techniques to predict the winning side based on the hero selection of the player and opponent, and the relationship between the selected heroes. Three traditional machine learning techniques, namely K-Nearest Neighbor, Logistic Regression, and Decision Tree are compared with ensemble learning with the most suitable parameters. The algorithms are evaluated using K-Fold Cross Validation, and the accuracy of each algorithm is measured. The results show that the victory prediction can be improved by considering the relationship between the selected heroes. In addition, ensemble learning is also competitive with traditional learning.