Smart Reckoning: Reducing the traffic of online multiplayer games using machine learning for movement prediction

Abstract

Massively Multiplayer Online Game (MMOG) players maintain consistent views of the positions of each other by periodically exchanging messages. Besides the fact that these messages can suffer delays that cause rendering inconsistencies, they also represent an overhead on the network. This overhead can be significant, as the number of MMOG players is very large, but reducing the number of messages is not trivial. The classic strategy to predict movement avoiding message exchange is based on the Dead Reckoning algorithm, which has several limitations. Other strategies have been proposed more recently that improve the results, but rely on expert knowledge. In this work we propose Smart Reckoning, a movement prediction strategy based on machine learning. The strategy consists of two phases. In the first phase, a learning model classifies whether the classical Dead Reckoning algorithm is able to predict the new avatar position correctly or not. In case the conclusion is negative, another learning model is used to predict the new direction. The proposed strategy was applied to the World of Warcraft game. The learning models were implemented with the Weka tool using real game trace data, and results are presented for the accuracy of multiple algorithms.