Integrating Machine Learning and Clinical Expertise: A Comparative Study for Improved Breast Cancer Diagnosis

Breast cancer affects women worldwide with high fatality rates. A critical initial phase in the rehabilitation treatment procedure is an accurate and early diagnosis. Due to many unknowns, mammography abnormality detection is a difficult task. By developing tools to help physicians identify and treat breast cancer, machine-learning approaches can dramatically increase patient survival rates. In this study focuses on four specific ML algorithms: K-Nearest Neighbors (KNN), Decision Tree, Random Forest and Support Vector Classifier (SVC). This study also includes the creation of two hybrid models: one combining SVC with Random Forest and an ensemble model consisting of SVC, Gradient Boosting and Multilayer Perceptron (MLP) with a Logistic Regression meta-model. Hard voting was employed in the ensemble, where predictions from multiple classifiers were combined and the class with the majority vote was selected. Using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset, we measure the classification test specificity, accuracy and sensitivity values of these models. The ensemble model results with the highest test accuracy of approximately 97.36%. The results found in this study provide an overview of machine learning methods for breast cancer detection, highlighting their potential for improving diagnostic performance.