Difficulty Modelling in Mobile Puzzle Games: An Empirical Study on Different Methods to Combine Player Analytics and Simulated Data

Predicting the perceived difficulty of mobile puzzle game levels is a critical challenge for developers, particularly when balancing personalized player experiences with the introduction of new content. The primary objective of this research is to resolve the cold-start problem, where a lack of historical player data hinders the accurate assessment of new levels. By leveraging a combination of player analytics, intrinsic level attributes, and data generated by reinforcement learning agents, developers can more effectively estimate the number of attempts required for level completion.

The analysis evaluates three distinct regression methods—Factorization Machines, Random Forests, and Artificial Neural Networks—to determine their efficacy across various data availability scenarios. Factorization Machines demonstrate superior performance and interpretability when applied to existing content, making them ideal for personalized difficulty adjustments. Conversely, Artificial Neural Networks and Random Forests prove more robust in cold-start scenarios, providing consistent predictive accuracy when historical player data is absent. The integration of simulated agent data is identified as a vital component for enhancing these predictions, as it allows for a more nuanced understanding of level design beyond simple global averages.

While reinforcement learning agents remain a powerful tool for augmenting predictive models, they exhibit limitations in distinguishing between easy and medium difficulty tiers. Despite this, the research concludes that a hybrid approach—utilizing neural networks for generalized predictions and factorization machines for established content—offers the most reliable framework for difficulty modeling. By synthesizing these diverse data streams, game designers can optimize level progression and player retention, ensuring that content remains engaging across the entire lifecycle of a mobile puzzle game.