A machine learning project for Dota 2 – a multiplayer online battle arena (MOBA) game that can predict the probability of winning a match given the characters chosen.
The project explores a number of features include: heroes synergistic and antagonistic relationships, players past performance, match duration and pick order of heroes. These features were tested and compared between five different supervised machine learning algorithms.
A test accuracy of 70% were achieved through incorporating these new features into the base model. Furthermore, it was found that through limiting the data to matches under 30 minutes only, a 77% test accuracy was obtained
Addtionally, the project includes an interface built using the library Tkinter that can help players in choosing the optimal character that will give them the edge in a match.
The disseration was coded using the Pycharm IDE which handles importing Python files from other folders internally. To run without IDE, the following snippet codes need to be added to files that import from other folders:
import sys
sys.path.extend([WORKING_DIRECTORY, FILE_DIRECTORY])
where WORKING_DIRECTORY is the root folder of the project, FILE_DIRECTORY is the subfolder where the file to be run is contained.