Data-Driven Prediction of Game Frame Rates Using Linear Regression and Random Forest Models

Abstract

Game frame rate (FPS) is a critical indicator of user experience and hardware performance in modern video games. However, players often lack accessible tools for predicting FPS under different configurations without running costly benchmarks. This study proposes a lightweight prediction framework that quantifies GPU performance, game workload, resolution scaling, and technologies such as DLSS and Frame Generation into measurable indicators. A synthetic dataset of 1296 configuration combinations was generated to ensure coverage of diverse scenarios. Linear Regression was implemented as a baseline model, and Random Forest was introduced as an improvement, demonstrating higher accuracy and robustness in both training and testing. An interactive command-line tool was developed, allowing users to input GPU, game, and resolution information to obtain FPS predictions. Although based on synthetic data, the framework illustrates the feasibility of FPS prediction via data-driven modeling and provides a foundation for future extensions with real-world benchmarks, broader hardware and game coverage, and advanced AI models for intelligent decision support.