Age of Empires II ["AoE2"] is a real-time strategy game set in the middle-ages. Players have to build a base and command units with the goal of defeating each other in matches of up to 8 players. When participating in AoE2’s competitive online-multiplayer they are rated with an Elo rating system not unlike Chess. This rating gives a good idea of the current form of a player and the rating difference can be converted directly into probability distributions for the outcome of a match or series between players. The goal of this project is to give a model capable of predicting in a meaningful way the outcome of matches played on the one-versus-one random map ladder ["RM_1v1"]. To this end the project introduces linear models to obtain a win probability ["WP"] based on: A) only Elo rating B) rating, map and civilization ["civ"]. Subsequently, non-linear models are introduced which give the WP based on rating, map and civ. The game was designed for different strategies to work well with certain civs and for civs to have bonuses on specific maps and disadvantages on others, similarly players may favour a certain map due to their experience or personal play style. The goal of this project is to quantify the influence of the factors map and civ in the initial win probability before a match.
This project is based on the dataset:"Age of Empires II DE Match Data" which contains two csv files with 3.153.767 data points representing online-multiplayer matches and a one-to-many rela- tionship to 9.732.500 data points representing the participating players. Preprocessing includes pruning the dataset for 1v1 matches and decisions on how to deal with missing or incomplete data points. Another important consideration is that the civs are rebalanced (sometimes drastically) with a new game patch. For now the data was trained on the most recent patch in the dataset. The dataset can be found here: https://www.kaggle.com/jerkeeler/age-of-empires-ii-de-match-data