Dual model approach to Dota2 match outcome prediction. Utilizing logistic regression and Keras.
=======
A full writeup can be found here.
The project requires a handful of python packages. Install them using:
pip install -r requirements.txt
In addition this project utilizes JupyterLab which can be found at https://jupyter.org/.
Each of the following Notebook Files will walk you through step by step on
1. Only 60% accuracy? That is not much better than predicting that radiant always wins.* Yes, but due to the extreme number of variables and hero combinations that attribute to predicting the match outcome, much more data would be needed to achieve higher accuracy.3. Why don't you use only 6k+ games to train your model then get rich by betting on pro games?* Sadly pro games and high mmr games are extremely different. The accuracy would be very low due to the differences and therefor not viable.4. Why did you not use other statistics, such as XPM, GPM or itemization?* This project was an attempt to utilize only the drafting data to predict which team would win. While including XPM, GPM, and other related statistics would be extremely useful in predicting the winning team, it would be dependent on data not available at the start of a game.5. How many games do I need to achieve the best accuracy possible?* In this project I utilized around 6k matches. To consistently achieve a “high” accuracy up to a million matches would have to be analyzed. Organizations such as Dotabuff and OpenDota utilize this amount of data and are able to consistently predict match outcomes.