Statistics Analysis in Dota 2 : Optimising Hero Selection and Predicting Match Outcomes

Tan, Jasper Guan Yu (2025) Statistics Analysis in Dota 2 : Optimising Hero Selection and Predicting Match Outcomes. Final Year Project (Bachelor), Tunku Abdul Rahman University of Technology and Management.

Text RDS _Jasper Tan Guan Yu_Fulltext.pdf Restricted to Registered users only Download (6MB)

Abstract

This project presents a web-based system designed to assist Dota 2 players in making better decisions during the hero drafting phase. With over 120 heroes and numerous possible combinations, choosing effective team compositions and counter-picks is a challenging task, especially for casual or less experienced players. To address this, the system uses machine learning and data analysis techniques to provide hero recommendations and predict match outcomes based on user-selected drafts. The project integrates Word2Vec and K-Nearest Neighbours (KNN) for hero recommendations, focusing on synergy and counters, while XGBoost is used to predict the probability of a team's victory. These models are trained using historical match data obtained through the OpenDota API. The backend is developed using Python and Flask, and the frontend is built with React, offering users a responsive and user-friendly interface. Testing showed that the system provides accurate and relevant predictions, with XGBoost achieving a high test accuracy of 98.97%. The system successfully supports players by offering real-time, data-driven insights to enhance their drafting strategies. This project also highlights the practical use of machine learning in gaming and contributes a valuable tool for both the Dota 2 community and the growing field of esports analytics.