A Robust Machine Learning Framework for Breast Cancer Prediction: Integrating Hybrid Feature Selection and Ensemble Learning

This study introduces a thorough approach for detecting breast cancer by combining advanced preprocessing, feature selection, hyperparameter tuning, and ensemble learning techniques. The process commences by doing meticulous data cleansing and implementing SMOTE to rectify any disparities in class distribution. The Interquartile Range (IQR) approach is employed to handle outliers, hence ensuring the integrity of the data. The feature selection process utilizes the HybridFS (LASSO and Recursive Feature Elimination) hybrid technique, which combines the regularization of LASSO with the iterative elimination of RFE to discover important predictors. Hyperparameter optimization for a Random Forest Classifier using Random Search optimizes model parameters to 97.23% accuracy. The final model improves performance with ensemble learning with ADABOOST, CATBOOST, and XGBOOST. The stacking ensemble approach achieves top results with 97% accuracy, 95% precision, 95% recall, 95% F1 score, and 96% AUC score. These findings demonstrate the effectiveness of combining methods to improve model robustness and minimize shortcomings. The work shows that advanced machine learning can create extremely accurate and trustworthy breast cancer prognostic models, which has major clinical diagnostic implications