Machine Learning Approaches to Choose Heroes in Dota 2

• Main Authors: Iuliia Porokhnenko, Petr Polezhaev, Alexander Shukhman
• Format: Article
• Language: English
• Published: FRUCT 2019-04-01
• Series: Proceedings of the XXth Conference of Open Innovations Association FRUCT
• Subjects: machine learning; classification models; eSports; multiplayer online game
• Online Access: https://fruct.org/publications/fruct24/files/Por.pdf

The winning in the multiplayer online game Dota 2 for teams is a sum of many factors. One of the most significant of them is the right choice of heroes for the team. It is possible to predict a match result based on the chosen heroes for both teams. This paper considers different approaches to predicting results of a match using machine learning methods to solve the classification problem. The experimental comparison of predictive classification models was done, including the optimization of their hyperparameters. It showed that the best classification models are linear regression, linear support vector machine, as well as neural network with Softplus and Sigmoid activation functions. The fastest of them is the linear regression model, so it is best suited for practical implementation.