GDC 2025 Session Viewer
Machine Learning Summit: Developing a Stats-Based Anti-Cheat Framework for 'Rainbow Six Siege'
Bettina Hein (Data Scientist, Ubisoft Entertainment)
Pass Type: All Access Pass, Summits Pass - Get your pass now!
Track: Programming
Format: Lecture
Vault Recording: TBD
Audience Level: All
Cheating in multiplayer online games negatively impacts player perception and engagement. To address this, a cheater detection system using machine learning driven by in-game player stats was developed for Rainbow Six Siege, operational since early 2024. It robustly identifies players with extremely high in-game performances. This talk covers how to create a labeled dataset to train our model, process data to make it robust against trends and variability, and regularly update and validate our model with the help of moderators. Moving forward, the goal is to establish this as the standard method for all our upcoming competitive FPS games.
Takeaway
The speakers will demonstrate how to construct a classification pipeline without ground-truth data and share best practices to iteratively create a labeled training set, detect in-game player behavior, set an optimal decision threshold to reduce false positives, and collaborate with a moderation team to validate detections and integrate insights on new behavior patterns.
Intended Audience
The talk requires no specific background. Experience with ML or in detecting player behavior is advantageous.
The session covers training ML models without ground truth data and behavior detection with minimal false positives. We also discuss how to effectively collaborate with non-technical teams. It benefits audiences dealing with similar challenges.
