Deep Learning for Pattern Recognition in Movements of Game Entities
This project is a work in progress made for the Master in Game Technologies at the Breda University of Applied Sciences. The goal of my research was to use deep learning to find patterns in player movements that indicate the usage of particular playstyles. A slightly modified version of Unity's open source Tanks! game was used to conduct our experiment. A neural network was written in Keras to do the pattern recognition.
A detailed explanation of the methodology, results and discussion can be found in the thesis which you can download here. Alternatively, a shorter version was also published to Gamasutra and can be found here.
All the source code of the project can be found on Github. The modified Tanks! game can be downloaded here. Note that this game was used to collect the data for the experiment, and will do so automatically. The data it generates can directly be used as input for the neural network also available on Github