A genetic algorithm to optimize SMOTE and GAN ratios in class imbalanced datasets.

Class imbalance is one of the problem easily encountered in the fields of data analysis and machine learning. When there is an imbalance in learning dataset, machine learning models become biased and learn inaccurate classifiers. To resolve such data imbalance problems, a strategy that increases the volume of data of minority classes is often used by applying the synthetic minority oversampling technique (SMOTE). Furthermore, the use of generative adversarial networks (GANs) for data oversampling has recently become more common. This research used a genetic algorithm to search and optimize the combinations of oversampling ratios based on the SMOTE and GAN techniques. The case in which the proposed method was used was compared with the cases in which a single technique was used to train either the imbalanced data or oversampled data. From the results, it was established that the classifier that learned the oversampled data with the optimized ratio using the proposed method was superior in classification performance.