Intelligent Healthcare System for Automated Breast Cancer Diagnosis using Advanced Ensemble Learning and Optimized Feature Engineering

Breast cancer remains one of the leading causes of mortality among women worldwide, necessitating early and accurate detection systems. This paper presents a comprehensive intelligent healthcare framework that integrates ensemble learning methodologies with advanced feature engineering techniques for automated breast cancer diagnosis. Our novel approach combines multiple machine learning classifiers (Naive Bayes, SVM with RBF kernel, Random Forest, J48 decision trees, and k-Nearest Neighbors) using both voting and stacking ensemble strategies. Additionally, we implement an innovative three-phased feature engineering framework utilizing PKIDiscretize discretization coupled with WrapperSubsetEval feature selection. Experimental evaluation on three benchmark datasets (Wisconsin, BCDR-F03, and BCDR-D01) demonstrates significant performance improvements. The proposed ensemble voting approach achieves 83.02% accuracy compared to 81.25% for the best individual classifier, representing a 2.92% improvement. Feature engineering further enhances diagnostic accuracy by 2-4% while achieving 53-63% dimensionality reduction. The ensemble classifier achieves superior evaluation metrics with TPR of 0.939, FPR of 0.323, and AUC of 0.909, demonstrating enhanced clinical applicability for breast cancer detection systems[1][2].

Keywords: Breast cancer diagnosis, ensemble learning, feature engineering, machine learning, medical image analysis, classification, dimensionality reduction.