Comparative Study of Machine Learning Algorithms for Breast Cancer Diagnosis Classification: From Data to Clinical Decision-Making

This study systematically compares the application of four classic machine learning algorithms—logistic regression, k-nearest neighbors (KNN), support vector machines (SVM), and decision trees—in breast cancer diagnosis classification. Based on the Wisconsin Breast Cancer Diagnosis Dataset, it explores the performance characteristics and clinical value of the algorithms, providing references for medical AI-assisted diagnosis.

Breast cancer is the most common malignant tumor among women globally, and early accurate diagnosis is crucial. Traditional diagnosis relies on empirical judgment, which has subjective errors; AI can provide objective assistance by analyzing historical cases. The Wisconsin Dataset contains 30 nuclear morphological features and benign/malignant labels for 569 cases, serving as a reliable foundation for supervised learning.

• Logistic Regression: Linear classification that outputs the probability of malignancy. It has strong interpretability but assumes a linear relationship and poorly adapts to non-linear patterns;
• K-Nearest Neighbors (KNN): Makes decisions based on similarity, is non-parametric with no distribution assumptions, but has low computational efficiency and is sensitive to scaling;
• Support Vector Machines (SVM): Finds the optimal decision boundary, adapts to non-linear patterns via kernel tricks, has strong robustness but complex parameter tuning;
• Decision Trees: Uses rule-based hierarchical judgment, has high interpretability but is prone to overfitting.

• Data Preprocessing: Requires missing value handling and standardization (for KNN/SVM);
• Training-Test Split: Common methods include random splitting or k-fold cross-validation;
• Hyperparameter Tuning: Grid/random search combined with cross-validation;
• Evaluation Metrics: Focus on precision (misdiagnosis rate), recall (missed diagnosis rate), F1 score, AUC-ROC, etc. The cost of missed diagnosis is higher than that of misdiagnosis.

Inferences based on algorithm characteristics: SVM may perform best on small to medium datasets; logistic regression is robust and provides feature importance; KNN performance fluctuates with K values and scaling; a single decision tree is prone to overfitting; ensemble methods (e.g., random forests) have greater potential.

• Data Quality: Requires standardization of data from different sources, and annotation costs are high;
• Interpretability: Doctors need to understand the algorithm's basis; logistic regression/decision trees are better;
• Ethical Responsibility: Definition of error liability and patient right to know need to be standardized;
• Continuous Learning: Needs to adapt to medical advances to avoid forgetting;
• Human-Machine Collaboration: AI assists rather than replaces doctors, leveraging the strengths of both parties.

Practice Recommendations: Data exploration visualization, feature engineering, stratified cross-validation, model error analysis, probability calibration; Future Directions: Deep learning (CNN), multi-modal fusion, personalized risk assessment, federated learning (multi-center collaboration under privacy protection).

This study lays a foundation for AI applications in breast cancer diagnosis, emphasizing that in addition to algorithm accuracy, interpretability, robustness, and ethical compliance are equally important. In the future, interdisciplinary collaboration between computer scientists, clinicians, ethicists, etc., is needed to promote technology implementation for the benefit of patients.