Engineering Features to Improve Pass Prediction in Soccer Simulation 2D Games

Engineering Features to Improve Pass Prediction in Soccer Simulation 2D Games

7 Jan 2024 | Nader Zare, Mahtab Sarvmaili, Aref Sayareh, Omid Amini, Stan Matwin, Amilcar Soares
This paper addresses the improvement of pass prediction in Soccer Simulation 2D (SS2D) games using Deep Neural Networks (DNN) and Random Forest (RF). The authors propose an embedded Data Extractor module that records decision-making processes in real-time, generating training data for machine learning models. Four data sorting techniques are applied to prepare the training data, and the models are evaluated against six top teams from RoboCup 2019. The study evaluates the importance of different feature groups on pass prediction accuracy, finding that the presence of all features improves prediction accuracy by 5% to 10%. The "X" sorting methods enhance model performance against various opponents, and features related to the position of the ball holder are more important than other positional features. The research concludes with a discussion on future directions, including disabling full state mode and exploring other models like recurrent neural networks.This paper addresses the improvement of pass prediction in Soccer Simulation 2D (SS2D) games using Deep Neural Networks (DNN) and Random Forest (RF). The authors propose an embedded Data Extractor module that records decision-making processes in real-time, generating training data for machine learning models. Four data sorting techniques are applied to prepare the training data, and the models are evaluated against six top teams from RoboCup 2019. The study evaluates the importance of different feature groups on pass prediction accuracy, finding that the presence of all features improves prediction accuracy by 5% to 10%. The "X" sorting methods enhance model performance against various opponents, and features related to the position of the ball holder are more important than other positional features. The research concludes with a discussion on future directions, including disabling full state mode and exploring other models like recurrent neural networks.
Reach us at info@study.space
[slides] Engineering Features to Improve Pass Prediction in Soccer Simulation 2D Games | StudySpace