Machine-Learning Model Developed to Combat Video-Game Cheating

Any video-game player knows how frustrating it is to compete with cheaters, but many don't realize the economic and other impacts on the game and developer. It also seems like no matter what actions a developer takes, some individuals always find ways to cheat a game. This is why computer scientists at the University of Texas at Dallas have taken an artificial intelligence (AI) approach to combat these players. 

The research was published in IEEE Transactions on Dependable and Secure Computing on Aug. 3.

The researchers used the popular first-person shooter game Counter-Strike to develop the new approach, but it can be applied to any massively multiplayer online (MMO) game where a central server receives data traffic. 

Counter-Strike is one of the most popular first-person shooter games on the market, meaning players are always using software cheats. The game involves teams of players working together to counter terrorists through bomb diffusion, hostage rescue, and securing plant locations. Players can purchase more powerful weapons by earning in-game currency.

Md Shihabul Islam is a UT Dallas computer science doctoral student in the Erik Jonsson School of Engineering and Computer Science. Islam, who is a Counter-Strike player himself, was lead author of the study.

“Sometimes when you're playing against players who use cheats you can tell, but sometimes it may not be evident,” he said. “It's not fair to the other players.”

The Economic Impact

Many players may see cheating as just a way of ruining the fun for others, but there are many more implications. Players often leave a game due to this behavior, which can cause an economic impact for the developer. 

In esports, which is a rapidly growing industry bringing in around $1 billion in annual revenues, cheating is punished through sanctions against teams and players. These can include disqualification, forfeiting winnings, or an all-out ban. 

Challenges of Detecting Cheating

One of the significant challenges surrounding cheating in MMO games is that it often goes undetected. Important data from a player's computer to the game server is encrypted, which means cheating is often only detected after game logs are decrypted, and it's too late. This is why the team at UT Dallas developed an approach that does not involve decryption, but it analyzes encrypted data traffic in real-time. 

Dr. Latifur Khan is a computer science professor and director of the Big Data Analytics and Management Lab at UT Dallas. He is also one of the authors of the study. 

“Players who cheat send traffic in a different way,” said Khan. “We're trying to capture those characteristics.”

Analyzing Game Traffic to Detect Patterns

The team's study involved 20 students using three software cheats in the game, including an aimbot, speed hack, and wallhack. The researchers then analyzed game traffic to and from the server, which led to discovering certain patterns that identified cheating behavior. 

The researchers used the data to train a machine-learning algorithm capable of predicting cheating based on the patterns and features. After adjusting the statistical model, it could be applied to larger groups. One aspect of their approach is that the data traffic is sent to a graphics processing unit, which quickens the process and decreases the workload of the central processing unit in the main server.

According to Islam, other gaming companies could use the new approach with their own data, eventually training gaming software for their games. After this software detects cheating behavior, it could be remedied right away.

“After detection,” Khan said, “we can give a warning and gracefully kick the player out if they continue with the cheating during a fixed time interval.

“Our aim is to ensure that games like Counter-Strike remain fun and fair for all players.”