Técnicas de Machine Learning para estimação de faturamento com o jogo League of Legends

Abstract

This work aims to develop a predictive model capable of estimating Riot Games' annual revenue from the game League of Legends (LoL), using Machine Learning (ML) techniques and public data related to the content released in the game. LoL is one of the most popular electronic games in the world, operating on a free-to-play model that monetizes mainly through the sale of cosmetic items such as skins. Despite its commercial success, Riot Games does not disclose its financial data in detail, making it challenging to understand the economic impact of its content strategies. To achieve the proposed objective, historical data on champions, skins, and events was collected and organized from reliable public sources, such as League of Legends Fandom. Data pre-processing included transformation into feature vectors suitable for application in machine learning models. Linear Regression, Decision Tree and Random Forest models were implemented and compared, assessing their performance in predicting revenue. A dimensionality reduction technique, Principal Component Analysis (PCA), was applied to optimize the models and improve the interpretability of the results. The results indicated that the Random Forest model, especially when combined with normalization and PCA techniques, showed the best performance in predicting turnover. Despite the limitations arising from the use of revenue estimates obtained from news sites and forums, the models managed to capture general trends, offering valuable insights into the factors that influence Riot Games' revenue. The study demonstrates the feasibility of applying Machine Learning techniques to revenue forecasting in contexts with limited or partially unavailable data.