The integration of machine learning (ML) algorithms in economic research has grown significantly in recent years. This undergraduate thesis focuses on applying these algorithms to predict stock market risks, which are critical for informing decisions made by policymakers, financial institutions, and investors. This work explores the efficacy of five boosting algorithms in forecasting three risk measures based on the Bovespa index (IBOV) returns: the Conditional Value at Risk, the Standard Deviation of Returns, and the Maximum Drawdown. The study investigates the theory of boosting methods, including AdaBoost, Gradient Boosting, XGBoost, LightGBM, and CatBoost, and compares their performance based on model metrics such as RMSE, MAE, training time, prediction time, and feature importance. The research uses Bovespa index data from August 2006 to May 2023, applying feature selection and hyperparameter tuning with random search, evaluated through cross-validation and key metrics. Key findings indicate that (i) the Spearman correlation was the most effective for feature selection, (ii) market sentiment and technical indicators were the most impactful in model outcomes, and (iii) LightGBM was the best algorithm for general risk prediction, while Adaboost and Gradient Boosting were the best algorithms for specific risk measures.