<p>The imbalanced data classification is common in many fields. One commonly used strategy is oversampling, which aims to achieve a balanced dataset by increasing the number of minority class samples. However, traditional oversampling methods tend to use linear interpolation mechanisms to simulate the generation of new data samples. This data generation method is too simple and is not suitable for complex data feature spaces. Based on the above problems, this paper proposes an improved SMOTE algorithm based on generative adversarial networks named GAN-SMOTE, which uses generative adversarial networks to generate minority class samples. First, the k-nearest neighbor algorithm is used to obtain the neighboring points of each sample point. Then the sample points are randomly selected in turn, and the neighboring points of the sample points are used as the real data of the generative adversarial network. After that, through continuous confrontation between the generator and the classifier, data that conforms to the data distribution characteristics around the sample point will be obtained. Finally, merge the newly generated data samples with the original data set to construct a balanced data set, and then classify the new data set. Comparative experiments with other oversampling algorithms show that this method outperforms other oversampling algorithms in terms of AUC and MCC, indicating that GAN-SMOTE has a certain effect on the classification of imbalanced data. Especially for data sets with high imbalance and small data volume, the classification effect after oversampling is more obvious.</p>