Effective Bots' Detection for Online Smartphone Game Using Multilayer Perceptron Neural Networks

Abstract

Online smartphone game bots can cause unfair behaviors and even shorten the game's life cycle. The random forest algorithm in machine learning is a widely used solution to identify game bots through behavioral features. Although the random forest algorithm can exactly detect more definite game bot players, some players belonging to the gray area cannot be detected accurately. Therefore, this study collects players' data and extracts the features to build the multilayer perceptron, neural network model, for effectively detecting online smartphone game bots. This approach calculates each player's abnormal rate to judge game bots and is evaluated on the famous mobile online game. Based on these abnormal rates, we then use K means to cluster players and further define the gray area. In the experimental evaluation, the results demonstrate the proposed learning model has better performance, not only increasing the accuracy but also reducing the error rate as compared with the random forest model in the same players' dataset. Accordingly, the proposed learning model can detect bot players more accurately and is feasible for real online smartphone games.