Comparative Analysis of Decision Tree Algorithms for DotA 2 Match Outcome Prediction

This paper presents a comparative analysis of various decision tree algorithms applied to the task of predicting match outcomes in Defense of the Ancients 2, a complex multiplayer online battle arena game. The study utilizes data acquired from the OpenDota Application Programming Interface and Steam Application Programming Interface, encompassing a wide range of game-related features. The core objective is to evaluate the performance of several decision tree models, including Classification and Regression Trees, C4.5, Random Forest, Gradient Boosting, Adaptive Boosting, and Extra Trees Classifier. The methodology includes data preprocessing, feature engineering, and model training using varying dataset sizes. The models were evaluated using metrics such as accuracy score. Results demonstrate that ensemble methods, such as Random Forest, Extra Trees Classifier and Histogram-based Gradient Boosting, generally outperform single decision tree algorithms, with the Extra Trees Classifier showing particularly strong performance. Furthermore, the experiments show that Adaptive Boosting's performance can vary widely. The findings of this study offer valuable insights into the effectiveness of different decision tree models for complex game outcome prediction and provide a baseline for further research in this domain. The ability to predict outcomes in Defense of the Ancients 2 has implications for player strategy and e-sports analytics.