Hybrid Multi-Step SMOTE-ENN Algorithm for Enhanced Breast Cancer Machine Learning Classification.

Abstract

Early detection of breast cancer is essential to prevent and reduce patient death roll, cost of operation on a patient, and provide early awareness for quick treatment. Researchers have proposed various methods for diagnosing and preventing breast cancer in women. It’s identified that prediction accuracy highly depends on the size, quality, and distribution of the dataset class (balanced or unbalanced data class), considering the health sector most public data available including the breast cancer dataset is still imbalanced and those that are addressed using oversampling techniques could lead to the addition of noise in the source data. Hence this study aims to adopt a hybrid multi-step oversampling SMOTE-ENN (Synthetic Minority Oversample Technique and Edited Nearest Neighbor) to improve the quality of the breast cancer dataset and limit the possibility of noise in the data. As a result, this improves the dataset used to train the machine learning model, and 0.99% accuracy is achieved during model evaluation