Data Preprocessing for DES-KNN and Its Application to Imbalanced Medical Data Classification

Learning from imbalanced data is a vital challenge for pattern classification. We often face the imbalanced data in medical decision tasks where at least one of the classes is represented by only a very small minority of the available data. We propose a novel framework for training base classifiers and preparing the dynamic selection dataset (dsel) to integrate data preprocessing and dynamic ensemble selection (des) methods for imbalanced data classification. des-knn algorithm has been chosen as the des method and its modifications base on oversampled training and validations sets using smote are discussed. The proposed modifications have been evaluated based on computer experiments carried out on 15 medical datasets with various imbalance ratios. The results of experiments show that the proposed framework is very useful, especially for tasks characterized by the small imbalance ratio.