A data-driven approach for online adaptation of game difficulty

Abstract

Dynamic adaptation of games with the objective of catering to the individual players' level of standard is an emerging and challenging research area of artificial intelligence in digital game. In this paper, we propose a data-driven approach for dynamic adaptation of game scenario difficulties. The goal is to fit the performance of the player to the desired conditions set by the designer. To this end, the data on player's in-game performance and dynamic game states are utilized for making adaptation decisions. Trained artificial neural networks are used to capture the relationship between dynamic game state, player performance, adaptation decision and the resultant game difficulty. Based on the predicted difficulty, adaptation of both direction and magnitude can be performed more effectively. Experimental study on a training game application is presented to demonstrate the efficiency and stability of the proposed approach.