DOTA 2 Win Loss Prediction from Item and Hero Data with Machine Learning

Abstract

Video gaming has become a titan in the overall market over the past decade, culminating in an estimated worth almost 180 billion US dollars by 2021. Aside from its growing influence in the overall market, video games have also created a new competitive format called eSports, a format where highly skilled players of certain video games play against each other in a tournament to see who the most skilled are and win a prize at the end. ESports are just one of many reasons why people have become interested in the idea of being able to predict the outcome of any given match between players. In this study, We conducted research on the importance of certain factors in determining the win or loss of any given Defense of the Ancients 2, better known as DOTA 2, match. We found that Item and Hero choices play a large role in winning any given match. From this, we concluded that we would be able to predict a match’s outcome solely based off of these two factors and created models to predict the outcome of any given match. In this study, we will be employing the use of Decision Tree, Random Tree and XGBoost classifiers in order to create our models. In the end, the XGBoost model ended up being our best model, with an accuracy of roughly 93% which can predict an outcome in roughly one minute.