Machine Learning Model for Skill Level Prediction in Multiplayer Online Battle Arena

Accurately predicting player skill levels in the Multiplayer Online Battle Arena (MOBA) game, Defense of the Ancients 2 (DotA 2), is important for improving the overall gaming experience and ensuring fair competition. This study proposed machine learning models to improve skill prediction, which can help developers create better matchmaking systems. We used data from 2.7 million matches and employed models such as Random Forest, LightGBM, and Artificial Neural Networks (ANN). To effectively represent player behavior, we introduced new features related to player strategies, items purchased, and in-game movements. Additionally, we introduced a match splitting method to address dynamic gameplay phases. The ANN model with Equal Duration-Based splitting demonstrated the best performance, achieving a Mean Absolute Error of 0.60 and an R-squared value of 0.85. Hypothesis testing confirmed that the model’s prediction errors aligned with actual player rank fluctuations, thus confirmed that the models’ performance is acceptable and consistent with the natural progression and fluctuation of player ranks.