Efficient Deep Learning Bot Detection in Games Using Time Windows and Long Short-Term Memory (LSTM)

Bots in video games has been gaining the interest of industry as well as academia as a problem that has been enabled by the recent advances in deep learning and reinforcement learning. In turn several studies have attempted to establish bot detectors in various video games. In this article, we introduce a bot detection model that can implemented in real-time and provide feedback on whether a player that is being observed is a bot or human. The model uses a limited feature set and amount of time of observation in order to be small and generalize easily to other domains. We trained and tested our model in a series of replays for Starcraft: Brood War and have yielded a higher accuracy than past studies and a fraction of detection time.