Predicting Quality of Video Gaming Experience Using Global-Scale Telemetry Data and Federated Learning

TL;DR

This study addresses the problem of predicting a game's FPS performance on a given device by analyzing micro- and macro-level determinants on a global-scale telemetry dataset. It introduces a federated learning framework with per-player and per-game learnable kernels to predict FPS distributions, mitigating privacy concerns and cold-start issues. The approach achieves a mean Wasserstein distance of $W_1=0.469$ on 5-class FPS distributions and demonstrates strong fidelity with a 42-class distribution predictor ($MAE=0.0213$, Top-1=0.3473), while enabling privacy-preserving, plug-and-play deployment. The work contributes a globally representative dataset, a novel LKK-based FL predictor, and quantitative insights into how hardware, game characteristics, and national socio-economic factors shape FPS, with practical implications for players, developers, and platforms.

Abstract

Frames Per Second (FPS) significantly affects the gaming experience. Providing players with accurate FPS estimates prior to purchase benefits both players and game developers. However, we have a limited understanding of how to predict a game's FPS performance on a specific device. In this paper, we first conduct a comprehensive analysis of a wide range of factors that may affect game FPS on a global-scale dataset to identify the determinants of FPS. This includes player-side and game-side characteristics, as well as country-level socio-economic statistics. Furthermore, recognizing that accurate FPS predictions require extensive user data, which raises privacy concerns, we propose a federated learning-based model to ensure user privacy. Each player and game is assigned a unique learnable knowledge kernel that gradually extracts latent features for improved accuracy. We also introduce a novel training and prediction scheme that allows these kernels to be dynamically plug-and-play, effectively addressing cold start issues. To train this model with minimal bias, we collected a large telemetry dataset from 224 countries and regions, 100,000 users, and 835 games. Our model achieved a mean Wasserstein distance of 0.469 between predicted and ground truth FPS distributions, outperforming all baseline methods.

Figures (11)

• Figure 1: The FPS of a game running on a device affect player's performance and mood. By analyzing telemetry data from all over the world, a federated learning-based FPS predictor is proposed for the convenience of players and game makers.
• Figure 2: The result of the game performance benchmark tool provided by Black Myth: Wukong, an AAA game released in August 2024. 95% FPS floor is adopted as one of the main game performance metrics in this benchmark tool.
• Figure 3: Performance and configuration guide for the game "A Plague Tale: Requiem" on (a) Epic GameStore, (b) Steam, and (c) Xbox Games. All game platforms provide minimum and recommended system configuration, while Xbox Games additionally provides data-driven binary running ability estimation.
• Figure 4: The global distribution of player number (log scale) in this telemetry dataset. The United States (18.87%), Russia (6.98%), Brazil (5.57%), China (4.85%), and Germany (4.27%) make up the largest share of players.
• Figure 5: The distribution of 95% FPS floor across CPU and GPU core parameters. Higher-spec GPU and CPU directly lead to a higher 95% FPS floor.