Breast cancer (BC) is a major health concern affecting women worldwide, and early detection is crucial for effective treatment and improved survival rates. In this study, we propose a novel BC prediction framework based on iterative optimization with Bayesian hyperparameter tuning applied to the Wisconsin Diagnostic Breast Cancer (WDBC) and Surveillance, Epidemiology, and End Results (SEER) datasets. Our approach employs various Machine Learning (ML) algorithms, including tree-based, support vector machine (SVM)-based, K-nearest neighbor (KNN)-based, tree-based ensemble, and artificial neural network (ANN)-based ML models. The results demonstrated that the optimized models generally outperformed their non-optimized counterparts. Notably, the optimized AdaBoost model achieved a remarkable performance with 100% accuracy, precision, recall, and F1-score on the WDBC dataset. The optimized GentleBoost model exhibited a high performance of 95.3% accuracy, 97.4% precision, 93.1% recall, 95.2% F1-score, and 0.99 area under the curve (AUC) on the SEER dataset. These findings highlight the potential of our proposed framework for enhancing BC prediction accuracy and robustness, paving the way for future research and clinical application.