RoDAC: A Robust Data-centric Anti-Cheat Framework for Fair Online Competitive Gaming

Online gaming platforms face significant challenges from cheating behaviors that threaten fair play and user trust. In particular, first-person shooter (FPS) games are especially vulnerable to a wide range of sophisticated cheating techniques that directly undermine competitive balance. Among various cheating techniques in FPS games, this work specifically focuses on detecting Extra-Sensory Perception (ESP) cheats, which reveal hidden game information to users and are difficult to detect using traditional rule-based methods. We formulate the ESP cheat detection task as a binary classification problem that aims to distinguish between legitimate users and cheaters, using a tabular dataset extracted from user-match logs. In practice, the tabular dataset presents three key challenges, including label noise arising from mislabeling of legitimate users and cheaters, severe class imbalance between them, and distribution shifts caused by evolving user behaviors, all of which degrade model performance. To address these challenges, we propose RoDAC, a Robust Data-centric Anti-Cheat framework that sequentially applies tailored sample selection methods to mitigate label noise, class imbalance, and distribution shifts. We have deployed RoDAC in the PUBG: BATTLEGROUNDS competitive league and confirmed that the framework significantly improves detection accuracy and robustness compared to baseline methods. Our detailed analysis of computational efficiency and model compatibility highlights the practical benefits of our data-centric strategy for sustaining scalable and reliable anti-cheat systems in dynamic gaming environments.