Paramedical Sciences and Military Health

Introduction: Breast cancer is the second most common cause of death among women worldwide. Classification of different breast cancer subtypes is crucial for early diagnosis and individualized treatment. This study aims to develop predictive models of breast cancer subtypes using machine learning approaches applied to single-omic datasets.

Materials and Methods: RNA-seq expression datasets comprising 526 samples and miRNA expression data including 755 samples were obtained from the TCGA-BRCA Cancer Genome Atlas database. The models were trained using a comprehensive framework involving key feature extraction from single-omic data, integration of heterogeneous data sources, dimensionality reduction, and rigorous model evaluation. Performance was assessed using metrics such as accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC-ROC) across multiple classifiers including partial least squares regression, random forest, naive Bayes, decision tree, neural networks, and lasso regression.

Results: Random forest and lasso regression classifiers showed superior performance among all models tested, achieving accuracies of 92.81% and 95.01%, respectively. Their receiver operating characteristic (ROC) curves demonstrated an area under the curve (AUC) of 0.97. Moreover, high recall values (93.61% for random forest and 93.82% for lasso regression) indicate strong ability to correctly identify positive cases. Both classifiers also achieved high F1 scores, reflecting balanced precision and recall and overall robust predictive performance.

Conclusion: This study demonstrates the potential of AI-based predictive models in accurately classifying breast cancer subtypes. The proposed framework can be applied to available clinical data to support early detection, precise prognosis, and personalized treatment strategies, ultimately helping to reduce cancer-related morbidity and mortality.