The sustained increase in wine production across various regions of the world underscores the need to ensure consistent product quality to maintain competitiveness in a highly demanding global market. The evaluation of red wine quality has traditionally relied on sensory methods, which introduce subjectivity and variability. Therefore, it is essential to adopt more objective and efficient methods, such as chemical analysis, to ensure more consistent evaluations. Additionally, this study aims to reduce dependence on human experience and improve objectivity in wine classification by using Machine Learning algorithms. The methodology follows four phases: Dataset acquisition, Preprocessing, Model implementation with training models (LR, XGB, SVC, RF, GB, AD, BC, KNN, DT), and Model evaluation through metric tests (Accuracy, Precision, Recall, F1-Score, AUC_ROC, Avg. RMSE). The results show that the RF was the most effective, outperforming others across several metrics after the application of the SMOTE technique. With a precision of 0.9208, recall of 0.9474, F1-score of 0.8905, and accuracy of 0.9180, Random Forest achieved the lowest RMSE (28.15), indicating a lower error rate in the predictions.