Morphological Feature Selection and Classification of Breast Cancer Samples with Machine Learning Methods: Analytical Research

Objective: Breast cancer is one of the most common malignancies among women worldwide, and early detection is crucial for improving clinical outcomes. This study aims to classify benign and malignant breast tumors using machine learning approaches based on cytomorphological features and to identify the most influential variables contributing to classification performance. Material and Methods: The dataset, obtained from the Kaggle repository, comprised 699 patient records with 9 cytomorphological features scored from 1 to 10. Data preprocessing and feature selection procedures were applied prior to model development. Logistic regression, Random Forest, and XGBoost models were implemented, and their performances were evaluated using accuracy, sensitivity, F1-score, and receiver operating characteristic (ROC)-area under the curve (AUC) metrics. SHAP (SHapley Additive explanations) analysis was employed to interpret model predictions and assess feature-level contributions. Results: Feature importance and SHAP analyses identified Uniformity of Cell Size, Uniformity of Cell Shape, Bare Nuclei, and Bland Chromatin as the most influential predictors. These findings highlight the critical role of cellular and nuclear morphological abnormalities in malignancy discrimination and enhance the interpretability and clinical relevance of the models. Conclusion: Among the evaluated models, XGBoost achieved the best overall performance, demonstrating the highest accuracy, sensitivity, and ROCAUC values. This indicates that ensemble-based learning methods provide superior discriminative capability for breast tumor classification compared to alternative approaches.