Improved Machine Learning-Based Predictive Models for Breast Cancer Diagnosis.

Area of Science:

• Oncology
• Biomedical Informatics
• Machine Learning

Background:

• Breast cancer is the leading cause of cancer death among American women.
• Machine learning offers potential for earlier breast cancer detection.
• Evaluating the efficiency of diagnostic models remains a challenge.

Purpose of the Study:

• To propose data exploratory techniques (DET) for improving breast cancer diagnostic accuracy.
• To develop and evaluate four predictive models for classifying malignant and benign breast cancer tumors.
• To compare the performance of developed models with existing studies.

Main Methods:

• Implemented a four-layered DET: feature distribution, correlation, elimination, and hyperparameter optimization.
• Developed and applied four classifiers: Support Vector Machine (SVM), Logistic Regression (LR), K-Nearest Neighbors (KNN), and Ensemble Classifier (EC).
• Utilized Wisconsin Diagnostic Breast Cancer (WDBC) and Breast Cancer Coimbra Dataset (BCCD) datasets, assessed using confusion matrices and K-fold cross-validation.

Main Results:

• DET significantly improved the diagnostic capability of the predictive models.
• Polynomial SVM achieved 99.3% accuracy, LR 98.06%, EC 97.61%, and KNN 97.35% on the WDBC dataset.
• The study demonstrated improved accuracy compared to previous research.

Conclusions:

• The proposed DET enhances the accuracy of machine learning models for breast cancer diagnosis.
• The findings provide a practical framework for physicians to select effective models for tumor prognosis.
• This research contributes to earlier and more accurate breast cancer detection and management.