ML Matchmaking Models: Enhancing Prediction With Artificial Player Data
17:15 - 17:45
Matchmaking is the core functionality for on-line games and esports. It can determine player satisfaction, engagement level, and ultimately player retention. The majority of the modern matchmaking systems are based on data-driven mechanisms such as outcome prediction models. All those models need a lot of data for an accurate match-up prediction and classification. However, new games or games with substantial changes to the game mechanics, usually, do not provide the required amount of data for accurate prediction. In the article, we explore the possibility of enhancing natural player-generated data with artificial bot-generated data. In a comparative experimental study, we compare different data sets and test them for accurate classification of players. We use a Naive Bayesian classifier and a Siamese neural network for model training and testing. The Naive Bayesian model shows a statistically significant increase in accuracy for data set composed of player-generated data enhanced with bot-generated data. Both models give meaningful insights into the classification process and implications for further studies.