41Potential biomarkers for breast cancer identification using machine learning techniques

Breast cancer is the most common type of cancer in women. Identification of biomarkers is necessary for prevention and treatment of this disease. Using the voting mechanism, an ensemble of Random Forest Classifiers and Gradient Boosting Classifier was used to select the top 20 features in the Breast Cancer Gene Expression Profiles (METABRIC) dataset from Kaggle. The resultant top 20 features were Nottingham prognostic index, age at diagnosis, ER status, lymph nodes examined, cohort, radiotherapy, hormone therapy, neoplasm histologic grade, GATA3, EGFR, tumor stage, IGF1R, PR status, BCL2, CCNE1, tumor size, HIF1A, DTX3, CCND1, CHEK1. Three machine learning techniques KNN, SVM, and RF were used to evaluate the performance, and RF attained highest accuracy of 95 %.