A comparative study on breast cancer classification with stratified shuffle split and K-fold cross validation via ensembled machine learning

Abstract

In breast cancer, early diagnosis and treatment method hold paramount significance for the augmented survival rates. Through a comprehensive dataset including clinical and genomic information, this study assesses the diverse analytical techniques used in breast cancer classification by the employment of four different machine learning algorithms. There were notable differences in classification findings, emphasizing the necessity of using adept analytical tools to improve the accuracy of breast cancer classification. Among individual algorithms, LGBM has the highest F1 score of 99.2% and a remarkable accuracy of 98.9%. Ensembles comprising AdaBoost, GBM, and RGF outperformed individual techniques with an astonishing 99.5% accuracy. The best ensemble algorithms prioritize features like worst texture, worst concave points, mean concave points, and mean texture, crucial for the classification. The examination of the advantages of ensemble learning methods, which combine predictions from many classifiers to improve classification performance, is at the heart of this the study. In particular, it is revealed how the k-fold and stratified shuffle split cross-validation methods differ in the classification results, providing clinicians a thorough understanding of the clinical ramifications to decipher the complex facets of breast cancer classification and identify crucial tumor traits that can distinguish malignant from benign cases.