On Combining Feature Selection and Over-Sampling Techniques for Breast Cancer Prediction

Breast cancer prediction datasets are usually class imbalanced, where the number of data samples in malignant and benign patient classes significantly different. Over-sampling techniques can be used to re-balance construct more effective models. Moreover, some related studies have considered feature selection remove irrelevant features from for further performance improvement. However, since order combining over-sampling result different training sets model, it is unknown which performs better. In this paper, information gain (IG) genetic algorithm (GA) methods synthetic minority technique (SMOTE) combinations. The experimental results based on two breast show that combination outperform single usage either highly imbalanced datasets. particular, performing IG first SMOTE second better choice. For other with a small imbalance ratio smaller features, enough an model.