Class imbalance and large number of classes severely degrade the classifiers’ performance while increasing computational costs. Some approaches are either computationally expensive or oversimplify the problem into a binary problem. To tackle these issues, the objective was to develop a Blend of Ranked-sequential Algorithms Executed for boosting-based classification (BREX), a hybrid hierarchical model combining Machine Learning techniques with a deterministic validation Leave-One-Out. BREX was applied in the white wine dataset with extreme imbalance class (n = 4898, m = 12, 7-classes, IR = 439.6). Fifteen classifiers were chosen, ranked and executed by computational simplicity strategy. Weighted-average measures (Recall, Specificity, Balanced Accuracy (BA), Precision, F1-score, MCC) were computed for performance evaluation. In the first stage (Euclidean) 3018 instances (61.6%) were classified reaching Recall = 0.62, Specificity = 0.81, BA = 0.71, Precision = 0.62, F1-score = 0.61, and MCC = 0.43; meanwhile, at the final stage (XGBoost) 4777 instances (97.5%) were classified reaching near-optimal performance with 0.98, 0.99, 0.98, 0.97, 0.98, and 0.96, respectively. BREX outperformed top standalone algorithms: Random Forest (best independent performer with 3521 predicted and BA = 0.77), IB1 (3299 predicted), and XGBoost (3276 predicted), showing a 27.2% improvement in BA. Computationally, BREX completed classification in 2.85 min (27.9 patterns/second), significantly faster than Random Forest (38.7 min; 13.6 × slower). This demonstrates BREX’s dual advantage in predictive accuracy and efficiency. Our proposal introduces a novel sequential algorithmic complexity structure that optimizes resource usage, representing a breakthrough for robust classification in extreme imbalance scenarios with optimal speed-performance trade-offs.