Detection of MMORPG bots based on behavior analysis

Game bots, i.e., autoplaying game clients, are currently causing troubles to both game publishers and bona fide players of Massively Multiplayer Online Role-Playing Games (MMORPGs). Use of game bots leads to collapse of game balance, decrease of player satisfaction, and even retirement from game. To prevent this, in-game polices, played by actual human players or game masters, often roam around game zones and individually question suspicious players, which is obviously laborious and ineffective task. In contrast to other work on automatic detection of MMORPG game bots based on the window events such as keystrokes, the game traffic, and the CAPTCHA test, our research focuses on log typically recorded by game publishers for database rollback. In particular, our research is based on discrepancies in action frequencies and action types in the log between human and bot characters. We propose the bot-detection methodology consisting of two stages. In the first stage an unknown character will be classified as "bot" if its frequencies of particular actions are much higher than those of known human characters. In the second stage, the rest of characters will be classified by the support vector machine classifier based on their action types. We evaluate the proposed methodology using game log of a Korean MMORPG titled Cabal Online and confirm its effectiveness.