AI-enabled prediction of video game player performance using the data from heterogeneous sensors

Area of Science:

• Sports Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• The rapid growth of eSports necessitates advanced analytics and training tools for players.
• Current performance evaluation methods in eSports are often subjective or lack comprehensive data integration.
• There is a need for objective, data-driven insights into player performance dynamics.

Purpose of the Study:

• To develop and validate an Artificial Intelligence (AI) enabled system for predicting eSports player in-game performance.
• To utilize sensor data (physiological, environmental, smart chair) for performance prediction.
• To assess the efficacy of an attention mechanism in improving prediction generalization and feature interpretability.

Main Methods:

• Collected physiological, environmental, and smart chair sensor data from professional and amateur eSports players.
• Assessed player performance using in-game logs in a multiplayer setting.
• Employed a recurrent neural network with an attention mechanism to predict future player performance.
• Evaluated model performance using the Area Under the Receiver Operating Characteristic Curve (ROC AUC).

Main Results:

• The AI model achieved an ROC AUC score of 0.73 in predicting whether a player would perform better or worse in the next 240 seconds.
• The model demonstrated the ability to predict performance even for players whose data were not included in the training set.
• An attention mechanism was found to enhance network generalization and provide clear feature importance.

Conclusions:

• The developed AI solution effectively predicts eSports player performance using sensor data.
• The system offers valuable applications for professional teams and amateur players, including performance monitoring and as a learning tool.
• This data-driven approach represents a significant advancement in eSports analytics and player development.