A Research On Application Of Cheating Recognition In Online Game Based On Machine Learning

With the development of online game industry,game cheating,such as games plugin,cheating script,personate transmitter and mouse macro,has become increasingly rampant.It has seriously affected game balance and experience of players.Especially in shooting-dominated E-sports online games,cheating not only affects game experience of players,but also changes the results of game competitions.Among those cheating methods,the mouse macro implemented by hardware driven,which uses a series of macro programming scripts,is the most difficult object to identify and prevent.Cheating using mouse macro is mostly verified by human reporting and manual detection.Its attributes,which are difficult to detect from the perspective of procedure and message,also become the best cheating method used in this thesis.Machine learning is an interdisciplinary subject.The analysis method based on machine learning algorithm has gradually become the mainstream of data recognition and classification.Using machine learning to analyze the cheating data of online games has become the focus of major online game companies nowadays.However,the existing work of using machine learning to identify cheating behavior of online games is mostly based on actual game applications,and there is little theoretical framework to summarize.Based on the analysis of the data characteristics of various types of online games,this thesis studies and designs a complete prototype of online game cheating recognition system based on machine learning.Taking a shooting online game platform as an example,the feasibility and validity of the research results in practical application are verified,which is a further study of machine learning method for online game cheating identification.Research has laid the foundation.This thesis also proposes a data mapping and processing method for online games.Unlike most of the methods that use game attributes as input directly,this method is more suitable for mature machine learning methods and has better recognition effect.The data set extracted by this method can be further studied by relevant researchers.In this thesis,the PUBG game is simulated and official gun data is used,in order to pursue data authenticity.The data extraction method based on image information matrix mapping in this thesis draws lessons from the realization process of image recognition,and the result of recognition accuracy is outstanding.From the results of data PCA,the self-made FCM data set in this thesis is concise and feasible for classification.Compared with the role attribute data used by predecessors as input method,the data scale is smaller and easier to learn.From the comparison results of model accuracy,the classification recognition rate of ResNet model is higher than that of SVM model based on FCM data set in this thesis.The accuracy rates of 99.22% and 97.12% of ResNet model are much higher than 84.7% of the previous neural network model based on attribute data.This not only illustrates the logical integrity and feasibility of the machine learning research process in the field of cheating recognition proposed in this thesis,but also shows that the cheating recognition method in this thesis successfully identifies the mouse macro cheating phenomenon which is difficult to detect and saves the time and cost of manual detection.Finally,this thesis proposes several methods to improve the model accuracy and machine learning indicators based on data set classification and game operation strategy,which provides some suggestions for the future research of machine learning in the field of online game cheating recognition.