Comparison of Feature Selection Methods in Breast Cancer Microarray Data

Aim: We aim to predict metastasis in breast cancer patients with tree-based conventional machine learning algorithms and observe which feature selection methods is more effective related microarray data reducing the number of features. Material Methods: Feature methods, least squares absolute shrinkage (LASSO), Boruta maximum relevance-minimum redundancy (MRMR) statistical preprocessing steps were first applied before like Decision-tree, Extremely randomized trees Gradient Boosting Tree on data. Results: Microarray 54675 features (202 (101/101 with/without metastases)) was reduced 235 features, then most important found algorithms. It observed that highest recall F-measure values obtained from XGBoost method precision value received Extra-tree method. The 10 arrays out variable importance listed. Conclusion: accurate results preprocessed for Extra-trees Statistical would be enough analysis metastases predictions.