Evolutionary feature selection for imbalanced data

Data imbalance and high dimensionality are two of the biggest changes in machine learning. To address such issues, feature selection is one of the Data Mining techniques that has become the focus of many research works. This work introduces a novel evolutionary hybrid feature selection algorithm based on filter statistics calculation. We conducted an experimental study over 44 imbalanced datasets, to evaluate the performance of the proposed evolutionary feature selection algorithm by considering seven filter statistics in the classification of imbalanced data. As a result, we offer a ranking of the filter statistics that better suit the hybrid evolutionary approach proposed for feature selection of imbalanced data, thus contributing to improving the Nearest Neighbor classifiers.