Predicting the outcomes of football matches is a challenging problem due to the sport s dynamic and multifactorial nature. In the Brazilian context, the Campeonato Brasileiro Série A exhibits specific characteristics, such as a high level of competitive balance, regional influences, and performance variability throughout the season that make predictive modeling particularly difficult. This work proposes the development and evaluation of a machine learning pipeline aimed at predicting match outcomes in Série A. To conduct the study, a comprehensive dataset was built from historical sources, including match statistics, recent team performance indicators, head-to-head records, regional factors, and metrics related to the sports betting market. A systematic feature engineering process was applied to capture temporal and contextual patterns relevant to Brazilian football. Several supervised classification models were evaluated, including Logistic Regression, Naive Bayes, K-Nearest Neighbors, Support Vector Machines, Random Forest, Gradient Boosting, AdaBoost, Multilayer Perceptron, and XGBoost. The models were compared using metrics suitable for potentially imbalanced problems, such as accuracy, precision, recall, and F1-score, with emphasis on weighted and macro averages.