Handling Class Imbalance Problem using Oversampling Techniques for Breast Cancer Prediction
Samta Rani, Tanvir Ahmad, Sarfaraz Masood · 2023
As a primary cause of death, breast cancer requires the highest care. Using machine learning models, recent next-generation sequencing methods capable of recording gene expression data have been utilized successfully to diagnose breast cancer. But the class imbalance issue affected the diagnostic outcomes of classification models. This issue is the source of the diagnostic errors. The diagnostic results have an impact on the lives of patients. Using oversampling techniques, this issue is resolved by balancing the samples of minority and majority classes. This paper examines the performance of all eight classification models with and without oversampling techniques using different metrics like precision, recall, and accuracy. The gene expression dataset, which contains samples from four molecular subtypes of breast cancer and one class of normal (non-cancer) samples, is used. It is publicly accessible on the TCGA-BRCA portal. Also, this study explores five oversampling strategies (Random, SMOTE, Borderline-SMOTE, SVMSMOTE, and SMOTENC) and their impact on model performance as a result of how they handle data imbalance problems.