Breast Cancer Classification: In-depth Exploration of Different Paradigms and Ensemble Technique

Many lives are lost to breast cancer every year. Prognosis and timely identification of cancer types are vital components of cancer research, as early diagnosis is critical to effective treatment. This study presents a comprehensive investigation into breast cancer classification, employing an extensive suite of datasets from Wisconsin Breast Cancer for machine learning models. Rigorous preprocessing techniques were applied to refine the dataset, segmented into groups of testing and training. The ensemble of models, comprising Logistic Regression, Decision Tree, Random Forest, XGBoost, Support Vector Machine (SVM), Gaussian Naive Bayes, K-Nearest Neighbors (KNN), AdaBoost, together with Deep Neural Network (DNN), underwent meticulous evaluation. For a more in-depth analysis, performance metrics like ROC-AUC scores, F1 Score, accuracy, precision, and recall were used. The ROC curve, or receiver operating characteristic provided visual insights, highlighting the trade-offs involving specificity and sensitivity. The ensemble model emerged as a promising approach, showcasing collaborative strengths across diverse algorithms, and achieves an accuracy of 98.2%. Confusion matrices and running time comparisons further elucidated model intricacies and computational efficiencies. This study will contribute valuable insights into breast cancer diagnostics, emphasizing the importance of tailored model selection and collaboration in clinical decision-making.