The Impact of Feature Selection on Different Machine Learning Models for Breast Cancer Classification

Abstract

Breast cancer appears to be a common type of cancer suffered by women globally, with considered high death rates. The survival rate of breast cancer patients decreases considerably for patients diagnosed at an advanced stage compared to those diagnosed at an early stage. The objective of this study is to investigate breast cancer classification and diagnosis task using the data from WBCD dataset. In our methodology, first, the breast cancer data was scaled. Then, four features selection methods were used to analyze the features. Pearson's Correlation method, Forward Selection method, Mutual Information and Univariate ROC-AUC were the used feature selectors. Next, different Machine Leaning models were applied including Support Vector Machine, Logistic Regression and XGBoost. Finally, the three models were cross-validated by 5-fold method. The ML models with different classifiers were evaluated based on several performance measures including accuracy, precision, recall, and F1-score. results show that Logistic Regression (LR) model with Forward Selection appeared to be the most successful classifier. The obtained classification accuracy, precision, and F1-score were 0.982, 0.983, 0.986; respectively. However, the highest recall score was 0.992 achieved by SVM model with Correlation feature selection. The developed model could potentially help the medical experts for the early diagnosis of breast cancer to decrease potential risk.