Predicting Game Engagement and Difficulty Using AI Players

TL;DR

This paper enhances automated game testing by combining Deep Reinforcement Learning with Monte Carlo Tree Search to better predict human player engagement and difficulty, especially in challenging game levels.

Contribution

It introduces an improved method that integrates DRL and MCTS, along with a new feature selection strategy, to more accurately predict player experience and game difficulty.

Findings

DRL+MCTS outperforms individual methods in predicting difficulty.

Enhanced feature selection improves prediction accuracy.

Best-case AI performance correlates more strongly with human data.

Abstract

This paper presents a novel approach to automated playtesting for the prediction of human player behavior and experience. It has previously been demonstrated that Deep Reinforcement Learning (DRL) game-playing agents can predict both game difficulty and player engagement, operationalized as average pass and churn rates. We improve this approach by enhancing DRL with Monte Carlo Tree Search (MCTS). We also motivate an enhanced selection strategy for predictor features, based on the observation that an AI agent's best-case performance can yield stronger correlations with human data than the agent's average performance. Both additions consistently improve the prediction accuracy, and the DRL-enhanced MCTS outperforms both DRL and vanilla MCTS in the hardest levels. We conclude that player modelling via automated playtesting can benefit from combining DRL and MCTS. Moreover, it can be worthwhile to investigate a subset of repeated best AI agent runs, if AI gameplay does not yield good predictions on average.