Financial data are often used in predicting users’ behaviors in business fields. The previous work usually focuses on the positive samples which means those specific persons can bring the profit to companies. However, in most cases, the proportion of positive samples is very small. The traditional algorithms do not perform well when the positive and negative samples are extremely unbalanced. To solve this problem, we propose an integrated network. Meanwhile, the original dataset includes both objective and index data, our method integrates the one-hot encoding and float encoding together to uniform the data which named hybrid encoding. Then, the article uses GAN framework to overcome the shortcoming of unbalanced dataset. Finally, voting rules put both data sensitive and data insensitive classifiers together to make a strong classifier. We evaluated the performance of our framework on a real world dataset and experimental results show that our method is effective, with 5%(±0.5%) to 87% improvement in accuracy as compared with other methods mentioned in this paper.