Machine learning for breast cancer diagnosis: A review of high-dimensional and class-imbalanced data approaches

Breast cancer remains a leading cause of mortality among women worldwide. Early and precise diagnosis is crucial for enhancing treatment outcomes and patient survival. However, early diagnosis is hindered by the high-dimensional, imbalanced, and noisy data. These challenges arise across gene expression, electronic health records, and medical imaging modalities. This paper explores recent advancements in Machine Learning (ML) approaches used to classify breast cancer, analysing dimensionality reduction, data balancing, and classification methods. The study identifies key technical challenges including overfitting, scalability limitations, class imbalance, and handling missing data, and evaluates how hybrid strategies mitigates them. The review emphasizes integrating preprocessing techniques with multilevel feature selection and ensemble modeling. These approaches consistently improve performance on diverse and high-dimensional datasets. The review also provides a step-by-step mapping of hybrid ML pipelines from datasets acquisition to deployment, offering practical insights into the diagnostic workflow. Based on 35 peer-reviewed studies, the review shows how hybrid models consistently outperform traditional approaches. Many achieve accuracy and recall above 95%. This review advocates for integrating explainable AI, standardized processing pipelines, and transparent validation to support clinical adoption. It also serves as a practical reference for developing robust, scalable ML-based breast cancer diagnostic systems, especially in resource-limited settings.