Wise Feature Selection for Breast Cancer Detection from a Clinical Dataset
Mahsa Bahrami, Mansour Vali · 2021
Breast cancer is a common cancer, especially in women. Early detection of breast cancer is taken an added importance because it alleviates the rate of mortality and facilitates treatment. Accurate automatic algorithms for the detection of breast cancer are needed. In this paper, we developed a number of different feature selection methods for accurate breast cancer detection. Clinical data was pre-processed and then wise feature selection based on feature importance was applied for feature selection. In addition to this, principal component analysis (PCA), incremental PCA, kernel PCA, independent component analysis, factor analysis, and singular value decomposition methods were implemented and analyzed for feature selection and dimension reduction. Finally, multi-layer perceptron was used for classification. The performance of feature selection methods was evaluated on Breast Cancer Wisconsin Diagnostic dataset with 569 recordings. The best accuracy, sensitivity, specificity, F1-score, and Cohen's kappa on the test data were 97.4%, 98.6%, 95.3%, 97.6%, and 0.94 respectively, with a feature importance method.